falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
2931 Commits
32 Branches
1012MB
Tree: a1e257b231
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a1e257b231'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2690 lines
80KB
Raw Blame History 

  1. /*

  2.  * MPEG Audio decoder

  3.  * Copyright (c) 2001, 2002 Fabrice Bellard.

  4.  *

  5.  * This library is free software; you can redistribute it and/or

  6.  * modify it under the terms of the GNU Lesser General Public

  7.  * License as published by the Free Software Foundation; either

  8.  * version 2 of the License, or (at your option) any later version.

  9.  *

  10.  * This library is distributed in the hope that it will be useful,

  11.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  12.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  13.  * Lesser General Public License for more details.

  14.  *

  15.  * You should have received a copy of the GNU Lesser General Public

  16.  * License along with this library; if not, write to the Free Software

  17.  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  18.  */

  19. 

  20. /**

  21.  * @file mpegaudiodec.c

  22.  * MPEG Audio decoder.

  23.  */ 

  24. 

  25. //#define DEBUG

  26. #include "avcodec.h"

  27. #include "mpegaudio.h"

  28. 

  29. /*

  30.  * TODO:

  31.  *  - in low precision mode, use more 16 bit multiplies in synth filter

  32.  *  - test lsf / mpeg25 extensively.

  33.  */

  34. 

  35. /* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg

  36.    audio decoder */

  37. #ifdef CONFIG_MPEGAUDIO_HP

  38. #define USE_HIGHPRECISION

  39. #endif

  40. 

  41. #ifdef USE_HIGHPRECISION

  42. #define FRAC_BITS   23   /* fractional bits for sb_samples and dct */

  43. #define WFRAC_BITS  16   /* fractional bits for window */

  44. #else

  45. #define FRAC_BITS   15   /* fractional bits for sb_samples and dct */

  46. #define WFRAC_BITS  14   /* fractional bits for window */

  47. #endif

  48. 

  49. #define FRAC_ONE    (1 << FRAC_BITS)

  50. 

  51. #define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)

  52. #define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))

  53. #define FIX(a)   ((int)((a) * FRAC_ONE))

  54. /* WARNING: only correct for posititive numbers */

  55. #define FIXR(a)   ((int)((a) * FRAC_ONE + 0.5))

  56. #define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)

  57. 

  58. #if FRAC_BITS <= 15

  59. typedef int16_t MPA_INT;

  60. #else

  61. typedef int32_t MPA_INT;

  62. #endif

  63. 

  64. /****************/

  65. 

  66. #define HEADER_SIZE 4

  67. #define BACKSTEP_SIZE 512

  68. 

  69. struct GranuleDef;

  70. 

  71. typedef struct MPADecodeContext {

  72.     uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE];	/* input buffer */

  73.     int inbuf_index;

  74.     uint8_t *inbuf_ptr, *inbuf;

  75.     int frame_size;

  76.     int free_format_frame_size; /* frame size in case of free format

  77.                                    (zero if currently unknown) */

  78.     /* next header (used in free format parsing) */

  79.     uint32_t free_format_next_header; 

  80.     int error_protection;

  81.     int layer;

  82.     int sample_rate;

  83.     int sample_rate_index; /* between 0 and 8 */

  84.     int bit_rate;

  85.     int old_frame_size;

  86.     GetBitContext gb;

  87.     int nb_channels;

  88.     int mode;

  89.     int mode_ext;

  90.     int lsf;

  91.     MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));

  92.     int synth_buf_offset[MPA_MAX_CHANNELS];

  93.     int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));

  94.     int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */

  95. #ifdef DEBUG

  96.     int frame_count;

  97. #endif

  98.     void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);

  99. } MPADecodeContext;

  100. 

  101. /* layer 3 "granule" */

  102. typedef struct GranuleDef {

  103.     uint8_t scfsi;

  104.     int part2_3_length;

  105.     int big_values;

  106.     int global_gain;

  107.     int scalefac_compress;

  108.     uint8_t block_type;

  109.     uint8_t switch_point;

  110.     int table_select[3];

  111.     int subblock_gain[3];

  112.     uint8_t scalefac_scale;

  113.     uint8_t count1table_select;

  114.     int region_size[3]; /* number of huffman codes in each region */

  115.     int preflag;

  116.     int short_start, long_end; /* long/short band indexes */

  117.     uint8_t scale_factors[40];

  118.     int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */

  119. } GranuleDef;

  120. 

  121. #define MODE_EXT_MS_STEREO 2

  122. #define MODE_EXT_I_STEREO  1

  123. 

  124. /* layer 3 huffman tables */

  125. typedef struct HuffTable {

  126.     int xsize;

  127.     const uint8_t *bits;

  128.     const uint16_t *codes;

  129. } HuffTable;

  130. 

  131. #include "mpegaudiodectab.h"

  132. 

  133. static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);

  134. static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);

  135. 

  136. /* vlc structure for decoding layer 3 huffman tables */

  137. static VLC huff_vlc[16]; 

  138. static uint8_t *huff_code_table[16];

  139. static VLC huff_quad_vlc[2];

  140. /* computed from band_size_long */

  141. static uint16_t band_index_long[9][23];

  142. /* XXX: free when all decoders are closed */

  143. #define TABLE_4_3_SIZE (8191 + 16)

  144. static int8_t  *table_4_3_exp;

  145. #if FRAC_BITS <= 15

  146. static uint16_t *table_4_3_value;

  147. #else

  148. static uint32_t *table_4_3_value;

  149. #endif

  150. /* intensity stereo coef table */

  151. static int32_t is_table[2][16];

  152. static int32_t is_table_lsf[2][2][16];

  153. static int32_t csa_table[8][4];

  154. static float csa_table_float[8][4];

  155. static int32_t mdct_win[8][36];

  156. 

  157. /* lower 2 bits: modulo 3, higher bits: shift */

  158. static uint16_t scale_factor_modshift[64];

  159. /* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */

  160. static int32_t scale_factor_mult[15][3];

  161. /* mult table for layer 2 group quantization */

  162. 

  163. #define SCALE_GEN(v) \

  164. { FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }

  165. 

  166. static int32_t scale_factor_mult2[3][3] = {

  167.     SCALE_GEN(4.0 / 3.0), /* 3 steps */

  168.     SCALE_GEN(4.0 / 5.0), /* 5 steps */

  169.     SCALE_GEN(4.0 / 9.0), /* 9 steps */

  170. };

  171. 

  172. /* 2^(n/4) */

  173. static uint32_t scale_factor_mult3[4] = {

  174.     FIXR(1.0),

  175.     FIXR(1.18920711500272106671),

  176.     FIXR(1.41421356237309504880),

  177.     FIXR(1.68179283050742908605),

  178. };

  179. 

  180. static MPA_INT window[512] __attribute__((aligned(16)));

  181.     

  182. /* layer 1 unscaling */

  183. /* n = number of bits of the mantissa minus 1 */

  184. static inline int l1_unscale(int n, int mant, int scale_factor)

  185. {

  186.     int shift, mod;

  187.     int64_t val;

  188. 

  189.     shift = scale_factor_modshift[scale_factor];

  190.     mod = shift & 3;

  191.     shift >>= 2;

  192.     val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);

  193.     shift += n;

  194.     /* NOTE: at this point, 1 <= shift >= 21 + 15 */

  195.     return (int)((val + (1LL << (shift - 1))) >> shift);

  196. }

  197. 

  198. static inline int l2_unscale_group(int steps, int mant, int scale_factor)

  199. {

  200.     int shift, mod, val;

  201. 

  202.     shift = scale_factor_modshift[scale_factor];

  203.     mod = shift & 3;

  204.     shift >>= 2;

  205. 

  206.     val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];

  207.     /* NOTE: at this point, 0 <= shift <= 21 */

  208.     if (shift > 0)

  209.         val = (val + (1 << (shift - 1))) >> shift;

  210.     return val;

  211. }

  212. 

  213. /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */

  214. static inline int l3_unscale(int value, int exponent)

  215. {

  216. #if FRAC_BITS <= 15    

  217.     unsigned int m;

  218. #else

  219.     uint64_t m;

  220. #endif

  221.     int e;

  222. 

  223.     e = table_4_3_exp[value];

  224.     e += (exponent >> 2);

  225.     e = FRAC_BITS - e;

  226. #if FRAC_BITS <= 15    

  227.     if (e > 31)

  228.         e = 31;

  229. #endif

  230.     m = table_4_3_value[value];

  231. #if FRAC_BITS <= 15    

  232.     m = (m * scale_factor_mult3[exponent & 3]);

  233.     m = (m + (1 << (e-1))) >> e;

  234.     return m;

  235. #else

  236.     m = MUL64(m, scale_factor_mult3[exponent & 3]);

  237.     m = (m + (uint64_t_C(1) << (e-1))) >> e;

  238.     return m;

  239. #endif

  240. }

  241. 

  242. /* all integer n^(4/3) computation code */

  243. #define DEV_ORDER 13

  244. 

  245. #define POW_FRAC_BITS 24

  246. #define POW_FRAC_ONE    (1 << POW_FRAC_BITS)

  247. #define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))

  248. #define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)

  249. 

  250. static int dev_4_3_coefs[DEV_ORDER];

  251. 

  252. static int pow_mult3[3] = {

  253.     POW_FIX(1.0),

  254.     POW_FIX(1.25992104989487316476),

  255.     POW_FIX(1.58740105196819947474),

  256. };

  257. 

  258. static void int_pow_init(void)

  259. {

  260.     int i, a;

  261. 

  262.     a = POW_FIX(1.0);

  263.     for(i=0;i<DEV_ORDER;i++) {

  264.         a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);

  265.         dev_4_3_coefs[i] = a;

  266.     }

  267. }

  268. 

  269. /* return the mantissa and the binary exponent */

  270. static int int_pow(int i, int *exp_ptr)

  271. {

  272.     int e, er, eq, j;

  273.     int a, a1;

  274.     

  275.     /* renormalize */

  276.     a = i;

  277.     e = POW_FRAC_BITS;

  278.     while (a < (1 << (POW_FRAC_BITS - 1))) {

  279.         a = a << 1;

  280.         e--;

  281.     }

  282.     a -= (1 << POW_FRAC_BITS);

  283.     a1 = 0;

  284.     for(j = DEV_ORDER - 1; j >= 0; j--)

  285.         a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);

  286.     a = (1 << POW_FRAC_BITS) + a1;

  287.     /* exponent compute (exact) */

  288.     e = e * 4;

  289.     er = e % 3;

  290.     eq = e / 3;

  291.     a = POW_MULL(a, pow_mult3[er]);

  292.     while (a >= 2 * POW_FRAC_ONE) {

  293.         a = a >> 1;

  294.         eq++;

  295.     }

  296.     /* convert to float */

  297.     while (a < POW_FRAC_ONE) {

  298.         a = a << 1;

  299.         eq--;

  300.     }

  301.     /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */

  302. #if POW_FRAC_BITS > FRAC_BITS

  303.     a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);

  304.     /* correct overflow */

  305.     if (a >= 2 * (1 << FRAC_BITS)) {

  306.         a = a >> 1;

  307.         eq++;

  308.     }

  309. #endif

  310.     *exp_ptr = eq;

  311.     return a;

  312. }

  313. 

  314. static int decode_init(AVCodecContext * avctx)

  315. {

  316.     MPADecodeContext *s = avctx->priv_data;

  317.     static int init=0;

  318.     int i, j, k;

  319. 

  320.     if (!init && !avctx->parse_only) {

  321.         /* scale factors table for layer 1/2 */

  322.         for(i=0;i<64;i++) {

  323.             int shift, mod;

  324.             /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  325.             shift = (i / 3);

  326.             mod = i % 3;

  327.             scale_factor_modshift[i] = mod | (shift << 2);

  328.         }

  329. 

  330.         /* scale factor multiply for layer 1 */

  331.         for(i=0;i<15;i++) {

  332.             int n, norm;

  333.             n = i + 2;

  334.             norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);

  335.             scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);

  336.             scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);

  337.             scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);

  338.             dprintf("%d: norm=%x s=%x %x %x\n",

  339.                     i, norm, 

  340.                     scale_factor_mult[i][0],

  341.                     scale_factor_mult[i][1],

  342.                     scale_factor_mult[i][2]);

  343.         }

  344.         

  345.         /* window */

  346.         /* max = 18760, max sum over all 16 coefs : 44736 */

  347.         for(i=0;i<257;i++) {

  348.             int v;

  349.             v = mpa_enwindow[i];

  350. #if WFRAC_BITS < 16

  351.             v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);

  352. #endif

  353.             window[i] = v;

  354.             if ((i & 63) != 0)

  355.                 v = -v;

  356.             if (i != 0)

  357.                 window[512 - i] = v;

  358.         }

  359.         

  360.         /* huffman decode tables */

  361.         huff_code_table[0] = NULL;

  362.         for(i=1;i<16;i++) {

  363.             const HuffTable *h = &mpa_huff_tables[i];

  364. 	    int xsize, x, y;

  365.             unsigned int n;

  366.             uint8_t *code_table;

  367. 

  368.             xsize = h->xsize;

  369.             n = xsize * xsize;

  370.             /* XXX: fail test */

  371.             init_vlc(&huff_vlc[i], 8, n, 

  372.                      h->bits, 1, 1, h->codes, 2, 2);

  373.             

  374.             code_table = av_mallocz(n);

  375.             j = 0;

  376.             for(x=0;x<xsize;x++) {

  377.                 for(y=0;y<xsize;y++)

  378.                     code_table[j++] = (x << 4) | y;

  379.             }

  380.             huff_code_table[i] = code_table;

  381.         }

  382.         for(i=0;i<2;i++) {

  383.             init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16, 

  384.                      mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1);

  385.         }

  386. 

  387.         for(i=0;i<9;i++) {

  388.             k = 0;

  389.             for(j=0;j<22;j++) {

  390.                 band_index_long[i][j] = k;

  391.                 k += band_size_long[i][j];

  392.             }

  393.             band_index_long[i][22] = k;

  394.         }

  395. 

  396. 	/* compute n ^ (4/3) and store it in mantissa/exp format */

  397. 	if (!av_mallocz_static(&table_4_3_exp,

  398. 			       TABLE_4_3_SIZE * sizeof(table_4_3_exp[0])))

  399. 	    return -1;

  400. 	if (!av_mallocz_static(&table_4_3_value,

  401. 			       TABLE_4_3_SIZE * sizeof(table_4_3_value[0])))

  402.             return -1;

  403.         

  404.         int_pow_init();

  405.         for(i=1;i<TABLE_4_3_SIZE;i++) {

  406.             int e, m;

  407.             m = int_pow(i, &e);

  408. #if 0

  409.             /* test code */

  410.             {

  411.                 double f, fm;

  412.                 int e1, m1;

  413.                 f = pow((double)i, 4.0 / 3.0);

  414.                 fm = frexp(f, &e1);

  415.                 m1 = FIXR(2 * fm);

  416. #if FRAC_BITS <= 15

  417.                 if ((unsigned short)m1 != m1) {

  418.                     m1 = m1 >> 1;

  419.                     e1++;

  420.                 }

  421. #endif

  422.                 e1--;

  423.                 if (m != m1 || e != e1) {

  424.                     printf("%4d: m=%x m1=%x e=%d e1=%d\n",

  425.                            i, m, m1, e, e1);

  426.                 }

  427.             }

  428. #endif

  429.             /* normalized to FRAC_BITS */

  430.             table_4_3_value[i] = m;

  431.             table_4_3_exp[i] = e;

  432.         }

  433.         

  434.         for(i=0;i<7;i++) {

  435.             float f;

  436.             int v;

  437.             if (i != 6) {

  438.                 f = tan((double)i * M_PI / 12.0);

  439.                 v = FIXR(f / (1.0 + f));

  440.             } else {

  441.                 v = FIXR(1.0);

  442.             }

  443.             is_table[0][i] = v;

  444.             is_table[1][6 - i] = v;

  445.         }

  446.         /* invalid values */

  447.         for(i=7;i<16;i++)

  448.             is_table[0][i] = is_table[1][i] = 0.0;

  449. 

  450.         for(i=0;i<16;i++) {

  451.             double f;

  452.             int e, k;

  453. 

  454.             for(j=0;j<2;j++) {

  455.                 e = -(j + 1) * ((i + 1) >> 1);

  456.                 f = pow(2.0, e / 4.0);

  457.                 k = i & 1;

  458.                 is_table_lsf[j][k ^ 1][i] = FIXR(f);

  459.                 is_table_lsf[j][k][i] = FIXR(1.0);

  460.                 dprintf("is_table_lsf %d %d: %x %x\n", 

  461.                         i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);

  462.             }

  463.         }

  464. 

  465.         if(avctx->antialias_algo == FF_AA_INT)

  466.             s->compute_antialias= compute_antialias_integer;

  467.         else

  468.             s->compute_antialias= compute_antialias_float;

  469.         for(i=0;i<8;i++) {

  470.             float ci, cs, ca;

  471.             ci = ci_table[i];

  472.             cs = 1.0 / sqrt(1.0 + ci * ci);

  473.             ca = cs * ci;

  474.             csa_table[i][0] = FIX(cs);

  475.             csa_table[i][1] = FIX(ca);

  476.             csa_table[i][2] = FIX(ca) + FIX(cs);

  477.             csa_table[i][3] = FIX(ca) - FIX(cs); 

  478.             csa_table_float[i][0] = cs;

  479.             csa_table_float[i][1] = ca;

  480.             csa_table_float[i][2] = ca + cs;

  481.             csa_table_float[i][3] = ca - cs; 

  482. //            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));

  483.         }

  484. 

  485.         /* compute mdct windows */

  486.         for(i=0;i<36;i++) {

  487.             int v;

  488.             v = FIXR(sin(M_PI * (i + 0.5) / 36.0));

  489.             mdct_win[0][i] = v;

  490.             mdct_win[1][i] = v;

  491.             mdct_win[3][i] = v;

  492.         }

  493.         for(i=0;i<6;i++) {

  494.             mdct_win[1][18 + i] = FIXR(1.0);

  495.             mdct_win[1][24 + i] = FIXR(sin(M_PI * ((i + 6) + 0.5) / 12.0));

  496.             mdct_win[1][30 + i] = FIXR(0.0);

  497. 

  498.             mdct_win[3][i] = FIXR(0.0);

  499.             mdct_win[3][6 + i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));

  500.             mdct_win[3][12 + i] = FIXR(1.0);

  501.         }

  502. 

  503.         for(i=0;i<12;i++)

  504.             mdct_win[2][i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));

  505.         

  506.         /* NOTE: we do frequency inversion adter the MDCT by changing

  507.            the sign of the right window coefs */

  508.         for(j=0;j<4;j++) {

  509.             for(i=0;i<36;i+=2) {

  510.                 mdct_win[j + 4][i] = mdct_win[j][i];

  511.                 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];

  512.             }

  513.         }

  514. 

  515. #if defined(DEBUG)

  516.         for(j=0;j<8;j++) {

  517.             printf("win%d=\n", j);

  518.             for(i=0;i<36;i++)

  519.                 printf("%f, ", (double)mdct_win[j][i] / FRAC_ONE);

  520.             printf("\n");

  521.         }

  522. #endif

  523.         init = 1;

  524.     }

  525. 

  526.     s->inbuf_index = 0;

  527.     s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];

  528.     s->inbuf_ptr = s->inbuf;

  529. #ifdef DEBUG

  530.     s->frame_count = 0;

  531. #endif

  532.     return 0;

  533. }

  534. 

  535. /* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */

  536. 

  537. /* cos(i*pi/64) */

  538. 

  539. #define COS0_0  FIXR(0.50060299823519630134)

  540. #define COS0_1  FIXR(0.50547095989754365998)

  541. #define COS0_2  FIXR(0.51544730992262454697)

  542. #define COS0_3  FIXR(0.53104259108978417447)

  543. #define COS0_4  FIXR(0.55310389603444452782)

  544. #define COS0_5  FIXR(0.58293496820613387367)

  545. #define COS0_6  FIXR(0.62250412303566481615)

  546. #define COS0_7  FIXR(0.67480834145500574602)

  547. #define COS0_8  FIXR(0.74453627100229844977)

  548. #define COS0_9  FIXR(0.83934964541552703873)

  549. #define COS0_10 FIXR(0.97256823786196069369)

  550. #define COS0_11 FIXR(1.16943993343288495515)

  551. #define COS0_12 FIXR(1.48416461631416627724)

  552. #define COS0_13 FIXR(2.05778100995341155085)

  553. #define COS0_14 FIXR(3.40760841846871878570)

  554. #define COS0_15 FIXR(10.19000812354805681150)

  555. 

  556. #define COS1_0 FIXR(0.50241928618815570551)

  557. #define COS1_1 FIXR(0.52249861493968888062)

  558. #define COS1_2 FIXR(0.56694403481635770368)

  559. #define COS1_3 FIXR(0.64682178335999012954)

  560. #define COS1_4 FIXR(0.78815462345125022473)

  561. #define COS1_5 FIXR(1.06067768599034747134)

  562. #define COS1_6 FIXR(1.72244709823833392782)

  563. #define COS1_7 FIXR(5.10114861868916385802)

  564. 

  565. #define COS2_0 FIXR(0.50979557910415916894)

  566. #define COS2_1 FIXR(0.60134488693504528054)

  567. #define COS2_2 FIXR(0.89997622313641570463)

  568. #define COS2_3 FIXR(2.56291544774150617881)

  569. 

  570. #define COS3_0 FIXR(0.54119610014619698439)

  571. #define COS3_1 FIXR(1.30656296487637652785)

  572. 

  573. #define COS4_0 FIXR(0.70710678118654752439)

  574. 

  575. /* butterfly operator */

  576. #define BF(a, b, c)\

  577. {\

  578.     tmp0 = tab[a] + tab[b];\

  579.     tmp1 = tab[a] - tab[b];\

  580.     tab[a] = tmp0;\

  581.     tab[b] = MULL(tmp1, c);\

  582. }

  583. 

  584. #define BF1(a, b, c, d)\

  585. {\

  586.     BF(a, b, COS4_0);\

  587.     BF(c, d, -COS4_0);\

  588.     tab[c] += tab[d];\

  589. }

  590. 

  591. #define BF2(a, b, c, d)\

  592. {\

  593.     BF(a, b, COS4_0);\

  594.     BF(c, d, -COS4_0);\

  595.     tab[c] += tab[d];\

  596.     tab[a] += tab[c];\

  597.     tab[c] += tab[b];\

  598.     tab[b] += tab[d];\

  599. }

  600. 

  601. #define ADD(a, b) tab[a] += tab[b]

  602. 

  603. /* DCT32 without 1/sqrt(2) coef zero scaling. */

  604. static void dct32(int32_t *out, int32_t *tab)

  605. {

  606.     int tmp0, tmp1;

  607. 

  608.     /* pass 1 */

  609.     BF(0, 31, COS0_0);

  610.     BF(1, 30, COS0_1);

  611.     BF(2, 29, COS0_2);

  612.     BF(3, 28, COS0_3);

  613.     BF(4, 27, COS0_4);

  614.     BF(5, 26, COS0_5);

  615.     BF(6, 25, COS0_6);

  616.     BF(7, 24, COS0_7);

  617.     BF(8, 23, COS0_8);

  618.     BF(9, 22, COS0_9);

  619.     BF(10, 21, COS0_10);

  620.     BF(11, 20, COS0_11);

  621.     BF(12, 19, COS0_12);

  622.     BF(13, 18, COS0_13);

  623.     BF(14, 17, COS0_14);

  624.     BF(15, 16, COS0_15);

  625. 

  626.     /* pass 2 */

  627.     BF(0, 15, COS1_0);

  628.     BF(1, 14, COS1_1);

  629.     BF(2, 13, COS1_2);

  630.     BF(3, 12, COS1_3);

  631.     BF(4, 11, COS1_4);

  632.     BF(5, 10, COS1_5);

  633.     BF(6,  9, COS1_6);

  634.     BF(7,  8, COS1_7);

  635.     

  636.     BF(16, 31, -COS1_0);

  637.     BF(17, 30, -COS1_1);

  638.     BF(18, 29, -COS1_2);

  639.     BF(19, 28, -COS1_3);

  640.     BF(20, 27, -COS1_4);

  641.     BF(21, 26, -COS1_5);

  642.     BF(22, 25, -COS1_6);

  643.     BF(23, 24, -COS1_7);

  644.     

  645.     /* pass 3 */

  646.     BF(0, 7, COS2_0);

  647.     BF(1, 6, COS2_1);

  648.     BF(2, 5, COS2_2);

  649.     BF(3, 4, COS2_3);

  650.     

  651.     BF(8, 15, -COS2_0);

  652.     BF(9, 14, -COS2_1);

  653.     BF(10, 13, -COS2_2);

  654.     BF(11, 12, -COS2_3);

  655.     

  656.     BF(16, 23, COS2_0);

  657.     BF(17, 22, COS2_1);

  658.     BF(18, 21, COS2_2);

  659.     BF(19, 20, COS2_3);

  660.     

  661.     BF(24, 31, -COS2_0);

  662.     BF(25, 30, -COS2_1);

  663.     BF(26, 29, -COS2_2);

  664.     BF(27, 28, -COS2_3);

  665. 

  666.     /* pass 4 */

  667.     BF(0, 3, COS3_0);

  668.     BF(1, 2, COS3_1);

  669.     

  670.     BF(4, 7, -COS3_0);

  671.     BF(5, 6, -COS3_1);

  672.     

  673.     BF(8, 11, COS3_0);

  674.     BF(9, 10, COS3_1);

  675.     

  676.     BF(12, 15, -COS3_0);

  677.     BF(13, 14, -COS3_1);

  678.     

  679.     BF(16, 19, COS3_0);

  680.     BF(17, 18, COS3_1);

  681.     

  682.     BF(20, 23, -COS3_0);

  683.     BF(21, 22, -COS3_1);

  684.     

  685.     BF(24, 27, COS3_0);

  686.     BF(25, 26, COS3_1);

  687.     

  688.     BF(28, 31, -COS3_0);

  689.     BF(29, 30, -COS3_1);

  690.     

  691.     /* pass 5 */

  692.     BF1(0, 1, 2, 3);

  693.     BF2(4, 5, 6, 7);

  694.     BF1(8, 9, 10, 11);

  695.     BF2(12, 13, 14, 15);

  696.     BF1(16, 17, 18, 19);

  697.     BF2(20, 21, 22, 23);

  698.     BF1(24, 25, 26, 27);

  699.     BF2(28, 29, 30, 31);

  700.     

  701.     /* pass 6 */

  702.     

  703.     ADD( 8, 12);

  704.     ADD(12, 10);

  705.     ADD(10, 14);

  706.     ADD(14,  9);

  707.     ADD( 9, 13);

  708.     ADD(13, 11);

  709.     ADD(11, 15);

  710. 

  711.     out[ 0] = tab[0];

  712.     out[16] = tab[1];

  713.     out[ 8] = tab[2];

  714.     out[24] = tab[3];

  715.     out[ 4] = tab[4];

  716.     out[20] = tab[5];

  717.     out[12] = tab[6];

  718.     out[28] = tab[7];

  719.     out[ 2] = tab[8];

  720.     out[18] = tab[9];

  721.     out[10] = tab[10];

  722.     out[26] = tab[11];

  723.     out[ 6] = tab[12];

  724.     out[22] = tab[13];

  725.     out[14] = tab[14];

  726.     out[30] = tab[15];

  727.     

  728.     ADD(24, 28);

  729.     ADD(28, 26);

  730.     ADD(26, 30);

  731.     ADD(30, 25);

  732.     ADD(25, 29);

  733.     ADD(29, 27);

  734.     ADD(27, 31);

  735. 

  736.     out[ 1] = tab[16] + tab[24];

  737.     out[17] = tab[17] + tab[25];

  738.     out[ 9] = tab[18] + tab[26];

  739.     out[25] = tab[19] + tab[27];

  740.     out[ 5] = tab[20] + tab[28];

  741.     out[21] = tab[21] + tab[29];

  742.     out[13] = tab[22] + tab[30];

  743.     out[29] = tab[23] + tab[31];

  744.     out[ 3] = tab[24] + tab[20];

  745.     out[19] = tab[25] + tab[21];

  746.     out[11] = tab[26] + tab[22];

  747.     out[27] = tab[27] + tab[23];

  748.     out[ 7] = tab[28] + tab[18];

  749.     out[23] = tab[29] + tab[19];

  750.     out[15] = tab[30] + tab[17];

  751.     out[31] = tab[31];

  752. }

  753. 

  754. #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)

  755. 

  756. #if FRAC_BITS <= 15

  757. 

  758. static inline int round_sample(int sum)

  759. {

  760.     int sum1;

  761.     sum1 = (sum + (1 << (OUT_SHIFT - 1))) >> OUT_SHIFT;

  762.     if (sum1 < -32768)

  763.         sum1 = -32768;

  764.     else if (sum1 > 32767)

  765.         sum1 = 32767;

  766.     return sum1;

  767. }

  768. 

  769. #if defined(ARCH_POWERPC_405)

  770. 

  771. /* signed 16x16 -> 32 multiply add accumulate */

  772. #define MACS(rt, ra, rb) \

  773.     asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));

  774. 

  775. /* signed 16x16 -> 32 multiply */

  776. #define MULS(ra, rb) \

  777.     ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; })

  778. 

  779. #else

  780. 

  781. /* signed 16x16 -> 32 multiply add accumulate */

  782. #define MACS(rt, ra, rb) rt += (ra) * (rb)

  783. 

  784. /* signed 16x16 -> 32 multiply */

  785. #define MULS(ra, rb) ((ra) * (rb))

  786. 

  787. #endif

  788. 

  789. #else

  790. 

  791. static inline int round_sample(int64_t sum) 

  792. {

  793.     int sum1;

  794.     sum1 = (int)((sum + (int64_t_C(1) << (OUT_SHIFT - 1))) >> OUT_SHIFT);

  795.     if (sum1 < -32768)

  796.         sum1 = -32768;

  797.     else if (sum1 > 32767)

  798.         sum1 = 32767;

  799.     return sum1;

  800. }

  801. 

  802. #define MULS(ra, rb) MUL64(ra, rb)

  803. 

  804. #endif

  805. 

  806. #define SUM8(sum, op, w, p) \

  807. {                                               \

  808.     sum op MULS((w)[0 * 64], p[0 * 64]);\

  809.     sum op MULS((w)[1 * 64], p[1 * 64]);\

  810.     sum op MULS((w)[2 * 64], p[2 * 64]);\

  811.     sum op MULS((w)[3 * 64], p[3 * 64]);\

  812.     sum op MULS((w)[4 * 64], p[4 * 64]);\

  813.     sum op MULS((w)[5 * 64], p[5 * 64]);\

  814.     sum op MULS((w)[6 * 64], p[6 * 64]);\

  815.     sum op MULS((w)[7 * 64], p[7 * 64]);\

  816. }

  817. 

  818. #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \

  819. {                                               \

  820.     int tmp;\

  821.     tmp = p[0 * 64];\

  822.     sum1 op1 MULS((w1)[0 * 64], tmp);\

  823.     sum2 op2 MULS((w2)[0 * 64], tmp);\

  824.     tmp = p[1 * 64];\

  825.     sum1 op1 MULS((w1)[1 * 64], tmp);\

  826.     sum2 op2 MULS((w2)[1 * 64], tmp);\

  827.     tmp = p[2 * 64];\

  828.     sum1 op1 MULS((w1)[2 * 64], tmp);\

  829.     sum2 op2 MULS((w2)[2 * 64], tmp);\

  830.     tmp = p[3 * 64];\

  831.     sum1 op1 MULS((w1)[3 * 64], tmp);\

  832.     sum2 op2 MULS((w2)[3 * 64], tmp);\

  833.     tmp = p[4 * 64];\

  834.     sum1 op1 MULS((w1)[4 * 64], tmp);\

  835.     sum2 op2 MULS((w2)[4 * 64], tmp);\

  836.     tmp = p[5 * 64];\

  837.     sum1 op1 MULS((w1)[5 * 64], tmp);\

  838.     sum2 op2 MULS((w2)[5 * 64], tmp);\

  839.     tmp = p[6 * 64];\

  840.     sum1 op1 MULS((w1)[6 * 64], tmp);\

  841.     sum2 op2 MULS((w2)[6 * 64], tmp);\

  842.     tmp = p[7 * 64];\

  843.     sum1 op1 MULS((w1)[7 * 64], tmp);\

  844.     sum2 op2 MULS((w2)[7 * 64], tmp);\

  845. }

  846. 

  847. 

  848. /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:

  849.    32 samples. */

  850. /* XXX: optimize by avoiding ring buffer usage */

  851. static void synth_filter(MPADecodeContext *s1,

  852.                          int ch, int16_t *samples, int incr, 

  853.                          int32_t sb_samples[SBLIMIT])

  854. {

  855.     int32_t tmp[32];

  856.     register MPA_INT *synth_buf;

  857.     const register MPA_INT *w, *w2, *p;

  858.     int j, offset, v;

  859.     int16_t *samples2;

  860. #if FRAC_BITS <= 15

  861.     int sum, sum2;

  862. #else

  863.     int64_t sum, sum2;

  864. #endif

  865.     

  866.     dct32(tmp, sb_samples);

  867.     

  868.     offset = s1->synth_buf_offset[ch];

  869.     synth_buf = s1->synth_buf[ch] + offset;

  870. 

  871.     for(j=0;j<32;j++) {

  872.         v = tmp[j];

  873. #if FRAC_BITS <= 15

  874.         /* NOTE: can cause a loss in precision if very high amplitude

  875.            sound */

  876.         if (v > 32767)

  877.             v = 32767;

  878.         else if (v < -32768)

  879.             v = -32768;

  880. #endif

  881.         synth_buf[j] = v;

  882.     }

  883.     /* copy to avoid wrap */

  884.     memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));

  885. 

  886.     samples2 = samples + 31 * incr;

  887.     w = window;

  888.     w2 = window + 31;

  889. 

  890.     sum = 0;

  891.     p = synth_buf + 16;

  892.     SUM8(sum, +=, w, p);

  893.     p = synth_buf + 48;

  894.     SUM8(sum, -=, w + 32, p);

  895.     *samples = round_sample(sum);

  896.     samples += incr;

  897.     w++;

  898. 

  899.     /* we calculate two samples at the same time to avoid one memory

  900.        access per two sample */

  901.     for(j=1;j<16;j++) {

  902.         sum = 0;

  903.         sum2 = 0;

  904.         p = synth_buf + 16 + j;

  905.         SUM8P2(sum, +=, sum2, -=, w, w2, p);

  906.         p = synth_buf + 48 - j;

  907.         SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);

  908. 

  909.         *samples = round_sample(sum);

  910.         samples += incr;

  911.         *samples2 = round_sample(sum2);

  912.         samples2 -= incr;

  913.         w++;

  914.         w2--;

  915.     }

  916.     

  917.     p = synth_buf + 32;

  918.     sum = 0;

  919.     SUM8(sum, -=, w + 32, p);

  920.     *samples = round_sample(sum);

  921. 

  922.     offset = (offset - 32) & 511;

  923.     s1->synth_buf_offset[ch] = offset;

  924. }

  925. 

  926. /* cos(pi*i/24) */

  927. #define C1  FIXR(0.99144486137381041114)

  928. #define C3  FIXR(0.92387953251128675612)

  929. #define C5  FIXR(0.79335334029123516458)

  930. #define C7  FIXR(0.60876142900872063941)

  931. #define C9  FIXR(0.38268343236508977173)

  932. #define C11 FIXR(0.13052619222005159154)

  933. 

  934. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  935.    cases. */

  936. static void imdct12(int *out, int *in)

  937. {

  938.     int tmp;

  939.     int64_t in1_3, in1_9, in4_3, in4_9;

  940. 

  941.     in1_3 = MUL64(in[1], C3);

  942.     in1_9 = MUL64(in[1], C9);

  943.     in4_3 = MUL64(in[4], C3);

  944.     in4_9 = MUL64(in[4], C9);

  945.     

  946.     tmp = FRAC_RND(MUL64(in[0], C7) - in1_3 - MUL64(in[2], C11) + 

  947.                    MUL64(in[3], C1) - in4_9 - MUL64(in[5], C5));

  948.     out[0] = tmp;

  949.     out[5] = -tmp;

  950.     tmp = FRAC_RND(MUL64(in[0] - in[3], C9) - in1_3 + 

  951.                    MUL64(in[2] + in[5], C3) - in4_9);

  952.     out[1] = tmp;

  953.     out[4] = -tmp;

  954.     tmp = FRAC_RND(MUL64(in[0], C11) - in1_9 + MUL64(in[2], C7) -

  955.                    MUL64(in[3], C5) + in4_3 - MUL64(in[5], C1));

  956.     out[2] = tmp;

  957.     out[3] = -tmp;

  958.     tmp = FRAC_RND(MUL64(-in[0], C5) + in1_9 + MUL64(in[2], C1) + 

  959.                    MUL64(in[3], C11) - in4_3 - MUL64(in[5], C7));

  960.     out[6] = tmp;

  961.     out[11] = tmp;

  962.     tmp = FRAC_RND(MUL64(-in[0] + in[3], C3) - in1_9 + 

  963.                    MUL64(in[2] + in[5], C9) + in4_3);

  964.     out[7] = tmp;

  965.     out[10] = tmp;

  966.     tmp = FRAC_RND(-MUL64(in[0], C1) - in1_3 - MUL64(in[2], C5) -

  967.                    MUL64(in[3], C7) - in4_9 - MUL64(in[5], C11));

  968.     out[8] = tmp;

  969.     out[9] = tmp;

  970. }

  971. 

  972. #undef C1

  973. #undef C3

  974. #undef C5

  975. #undef C7

  976. #undef C9

  977. #undef C11

  978. 

  979. /* cos(pi*i/18) */

  980. #define C1 FIXR(0.98480775301220805936)

  981. #define C2 FIXR(0.93969262078590838405)

  982. #define C3 FIXR(0.86602540378443864676)

  983. #define C4 FIXR(0.76604444311897803520)

  984. #define C5 FIXR(0.64278760968653932632)

  985. #define C6 FIXR(0.5)

  986. #define C7 FIXR(0.34202014332566873304)

  987. #define C8 FIXR(0.17364817766693034885)

  988. 

  989. /* 0.5 / cos(pi*(2*i+1)/36) */

  990. static const int icos36[9] = {

  991.     FIXR(0.50190991877167369479),

  992.     FIXR(0.51763809020504152469),

  993.     FIXR(0.55168895948124587824),

  994.     FIXR(0.61038729438072803416),

  995.     FIXR(0.70710678118654752439),

  996.     FIXR(0.87172339781054900991),

  997.     FIXR(1.18310079157624925896),

  998.     FIXR(1.93185165257813657349),

  999.     FIXR(5.73685662283492756461),

  1000. };

  1001. 

  1002. static const int icos72[18] = {

  1003.     /* 0.5 / cos(pi*(2*i+19)/72) */

  1004.     FIXR(0.74009361646113053152),

  1005.     FIXR(0.82133981585229078570),

  1006.     FIXR(0.93057949835178895673),

  1007.     FIXR(1.08284028510010010928),

  1008.     FIXR(1.30656296487637652785),

  1009.     FIXR(1.66275476171152078719),

  1010.     FIXR(2.31011315767264929558),

  1011.     FIXR(3.83064878777019433457),

  1012.     FIXR(11.46279281302667383546),

  1013. 

  1014.     /* 0.5 / cos(pi*(2*(i + 18) +19)/72) */

  1015.     FIXR(-0.67817085245462840086),

  1016.     FIXR(-0.63023620700513223342),

  1017.     FIXR(-0.59284452371708034528),

  1018.     FIXR(-0.56369097343317117734),

  1019.     FIXR(-0.54119610014619698439),

  1020.     FIXR(-0.52426456257040533932),

  1021.     FIXR(-0.51213975715725461845),

  1022.     FIXR(-0.50431448029007636036),

  1023.     FIXR(-0.50047634258165998492),

  1024. };

  1025. 

  1026. /* using Lee like decomposition followed by hand coded 9 points DCT */

  1027. static void imdct36(int *out, int *in)

  1028. {

  1029.     int i, j, t0, t1, t2, t3, s0, s1, s2, s3;

  1030.     int tmp[18], *tmp1, *in1;

  1031.     int64_t in3_3, in6_6;

  1032. 

  1033.     for(i=17;i>=1;i--)

  1034.         in[i] += in[i-1];

  1035.     for(i=17;i>=3;i-=2)

  1036.         in[i] += in[i-2];

  1037. 

  1038.     for(j=0;j<2;j++) {

  1039.         tmp1 = tmp + j;

  1040.         in1 = in + j;

  1041. 

  1042.         in3_3 = MUL64(in1[2*3], C3);

  1043.         in6_6 = MUL64(in1[2*6], C6);

  1044. 

  1045.         tmp1[0] = FRAC_RND(MUL64(in1[2*1], C1) + in3_3 + 

  1046.                            MUL64(in1[2*5], C5) + MUL64(in1[2*7], C7));

  1047.         tmp1[2] = in1[2*0] + FRAC_RND(MUL64(in1[2*2], C2) + 

  1048.                                       MUL64(in1[2*4], C4) + in6_6 + 

  1049.                                       MUL64(in1[2*8], C8));

  1050.         tmp1[4] = FRAC_RND(MUL64(in1[2*1] - in1[2*5] - in1[2*7], C3));

  1051.         tmp1[6] = FRAC_RND(MUL64(in1[2*2] - in1[2*4] - in1[2*8], C6)) - 

  1052.             in1[2*6] + in1[2*0];

  1053.         tmp1[8] = FRAC_RND(MUL64(in1[2*1], C5) - in3_3 - 

  1054.                            MUL64(in1[2*5], C7) + MUL64(in1[2*7], C1));

  1055.         tmp1[10] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C8) - 

  1056.                                        MUL64(in1[2*4], C2) + in6_6 + 

  1057.                                        MUL64(in1[2*8], C4));

  1058.         tmp1[12] = FRAC_RND(MUL64(in1[2*1], C7) - in3_3 + 

  1059.                             MUL64(in1[2*5], C1) - 

  1060.                             MUL64(in1[2*7], C5));

  1061.         tmp1[14] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C4) + 

  1062.                                        MUL64(in1[2*4], C8) + in6_6 - 

  1063.                                        MUL64(in1[2*8], C2));

  1064.         tmp1[16] = in1[2*0] - in1[2*2] + in1[2*4] - in1[2*6] + in1[2*8];

  1065.     }

  1066. 

  1067.     i = 0;

  1068.     for(j=0;j<4;j++) {

  1069.         t0 = tmp[i];

  1070.         t1 = tmp[i + 2];

  1071.         s0 = t1 + t0;

  1072.         s2 = t1 - t0;

  1073. 

  1074.         t2 = tmp[i + 1];

  1075.         t3 = tmp[i + 3];

  1076.         s1 = MULL(t3 + t2, icos36[j]);

  1077.         s3 = MULL(t3 - t2, icos36[8 - j]);

  1078.         

  1079.         t0 = MULL(s0 + s1, icos72[9 + 8 - j]);

  1080.         t1 = MULL(s0 - s1, icos72[8 - j]);

  1081.         out[18 + 9 + j] = t0;

  1082.         out[18 + 8 - j] = t0;

  1083.         out[9 + j] = -t1;

  1084.         out[8 - j] = t1;

  1085.         

  1086.         t0 = MULL(s2 + s3, icos72[9+j]);

  1087.         t1 = MULL(s2 - s3, icos72[j]);

  1088.         out[18 + 9 + (8 - j)] = t0;

  1089.         out[18 + j] = t0;

  1090.         out[9 + (8 - j)] = -t1;

  1091.         out[j] = t1;

  1092.         i += 4;

  1093.     }

  1094. 

  1095.     s0 = tmp[16];

  1096.     s1 = MULL(tmp[17], icos36[4]);

  1097.     t0 = MULL(s0 + s1, icos72[9 + 4]);

  1098.     t1 = MULL(s0 - s1, icos72[4]);

  1099.     out[18 + 9 + 4] = t0;

  1100.     out[18 + 8 - 4] = t0;

  1101.     out[9 + 4] = -t1;

  1102.     out[8 - 4] = t1;

  1103. }

  1104. 

  1105. /* fast header check for resync */

  1106. static int check_header(uint32_t header)

  1107. {

  1108.     /* header */

  1109.     if ((header & 0xffe00000) != 0xffe00000)

  1110. 	return -1;

  1111.     /* layer check */

  1112.     if (((header >> 17) & 3) == 0)

  1113. 	return -1;

  1114.     /* bit rate */

  1115.     if (((header >> 12) & 0xf) == 0xf)

  1116. 	return -1;

  1117.     /* frequency */

  1118.     if (((header >> 10) & 3) == 3)

  1119. 	return -1;

  1120.     return 0;

  1121. }

  1122. 

  1123. /* header + layer + bitrate + freq + lsf/mpeg25 */

  1124. #define SAME_HEADER_MASK \

  1125.    (0xffe00000 | (3 << 17) | (0xf << 12) | (3 << 10) | (3 << 19))

  1126. 

  1127. /* header decoding. MUST check the header before because no

  1128.    consistency check is done there. Return 1 if free format found and

  1129.    that the frame size must be computed externally */

  1130. static int decode_header(MPADecodeContext *s, uint32_t header)

  1131. {

  1132.     int sample_rate, frame_size, mpeg25, padding;

  1133.     int sample_rate_index, bitrate_index;

  1134.     if (header & (1<<20)) {

  1135.         s->lsf = (header & (1<<19)) ? 0 : 1;

  1136.         mpeg25 = 0;

  1137.     } else {

  1138.         s->lsf = 1;

  1139.         mpeg25 = 1;

  1140.     }

  1141.     

  1142.     s->layer = 4 - ((header >> 17) & 3);

  1143.     /* extract frequency */

  1144.     sample_rate_index = (header >> 10) & 3;

  1145.     sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);

  1146.     sample_rate_index += 3 * (s->lsf + mpeg25);

  1147.     s->sample_rate_index = sample_rate_index;

  1148.     s->error_protection = ((header >> 16) & 1) ^ 1;

  1149.     s->sample_rate = sample_rate;

  1150. 

  1151.     bitrate_index = (header >> 12) & 0xf;

  1152.     padding = (header >> 9) & 1;

  1153.     //extension = (header >> 8) & 1;

  1154.     s->mode = (header >> 6) & 3;

  1155.     s->mode_ext = (header >> 4) & 3;

  1156.     //copyright = (header >> 3) & 1;

  1157.     //original = (header >> 2) & 1;

  1158.     //emphasis = header & 3;

  1159. 

  1160.     if (s->mode == MPA_MONO)

  1161.         s->nb_channels = 1;

  1162.     else

  1163.         s->nb_channels = 2;

  1164.     

  1165.     if (bitrate_index != 0) {

  1166.         frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];

  1167.         s->bit_rate = frame_size * 1000;

  1168.         switch(s->layer) {

  1169.         case 1:

  1170.             frame_size = (frame_size * 12000) / sample_rate;

  1171.             frame_size = (frame_size + padding) * 4;

  1172.             break;

  1173.         case 2:

  1174.             frame_size = (frame_size * 144000) / sample_rate;

  1175.             frame_size += padding;

  1176.             break;

  1177.         default:

  1178.         case 3:

  1179.             frame_size = (frame_size * 144000) / (sample_rate << s->lsf);

  1180.             frame_size += padding;

  1181.             break;

  1182.         }

  1183.         s->frame_size = frame_size;

  1184.     } else {

  1185.         /* if no frame size computed, signal it */

  1186.         if (!s->free_format_frame_size)

  1187.             return 1;

  1188.         /* free format: compute bitrate and real frame size from the

  1189.            frame size we extracted by reading the bitstream */

  1190.         s->frame_size = s->free_format_frame_size;

  1191.         switch(s->layer) {

  1192.         case 1:

  1193.             s->frame_size += padding  * 4;

  1194.             s->bit_rate = (s->frame_size * sample_rate) / 48000;

  1195.             break;

  1196.         case 2:

  1197.             s->frame_size += padding;

  1198.             s->bit_rate = (s->frame_size * sample_rate) / 144000;

  1199.             break;

  1200.         default:

  1201.         case 3:

  1202.             s->frame_size += padding;

  1203.             s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;

  1204.             break;

  1205.         }

  1206.     }

  1207.     

  1208. #if defined(DEBUG)

  1209.     printf("layer%d, %d Hz, %d kbits/s, ",

  1210.            s->layer, s->sample_rate, s->bit_rate);

  1211.     if (s->nb_channels == 2) {

  1212.         if (s->layer == 3) {

  1213.             if (s->mode_ext & MODE_EXT_MS_STEREO)

  1214.                 printf("ms-");

  1215.             if (s->mode_ext & MODE_EXT_I_STEREO)

  1216.                 printf("i-");

  1217.         }

  1218.         printf("stereo");

  1219.     } else {

  1220.         printf("mono");

  1221.     }

  1222.     printf("\n");

  1223. #endif

  1224.     return 0;

  1225. }

  1226. 

  1227. /* useful helper to get mpeg audio stream infos. Return -1 if error in

  1228.    header, otherwise the coded frame size in bytes */

  1229. int mpa_decode_header(AVCodecContext *avctx, uint32_t head)

  1230. {

  1231.     MPADecodeContext s1, *s = &s1;

  1232. 

  1233.     if (check_header(head) != 0)

  1234.         return -1;

  1235. 

  1236.     if (decode_header(s, head) != 0) {

  1237.         return -1;

  1238.     }

  1239. 

  1240.     switch(s->layer) {

  1241.     case 1:

  1242.         avctx->frame_size = 384;

  1243.         break;

  1244.     case 2:

  1245.         avctx->frame_size = 1152;

  1246.         break;

  1247.     default:

  1248.     case 3:

  1249.         if (s->lsf)

  1250.             avctx->frame_size = 576;

  1251.         else

  1252.             avctx->frame_size = 1152;

  1253.         break;

  1254.     }

  1255. 

  1256.     avctx->sample_rate = s->sample_rate;

  1257.     avctx->channels = s->nb_channels;

  1258.     avctx->bit_rate = s->bit_rate;

  1259.     avctx->sub_id = s->layer;

  1260.     return s->frame_size;

  1261. }

  1262. 

  1263. /* return the number of decoded frames */

  1264. static int mp_decode_layer1(MPADecodeContext *s)

  1265. {

  1266.     int bound, i, v, n, ch, j, mant;

  1267.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  1268.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  1269. 

  1270.     if (s->mode == MPA_JSTEREO) 

  1271.         bound = (s->mode_ext + 1) * 4;

  1272.     else

  1273.         bound = SBLIMIT;

  1274. 

  1275.     /* allocation bits */

  1276.     for(i=0;i<bound;i++) {

  1277.         for(ch=0;ch<s->nb_channels;ch++) {

  1278.             allocation[ch][i] = get_bits(&s->gb, 4);

  1279.         }

  1280.     }

  1281.     for(i=bound;i<SBLIMIT;i++) {

  1282.         allocation[0][i] = get_bits(&s->gb, 4);

  1283.     }

  1284. 

  1285.     /* scale factors */

  1286.     for(i=0;i<bound;i++) {

  1287.         for(ch=0;ch<s->nb_channels;ch++) {

  1288.             if (allocation[ch][i])

  1289.                 scale_factors[ch][i] = get_bits(&s->gb, 6);

  1290.         }

  1291.     }

  1292.     for(i=bound;i<SBLIMIT;i++) {

  1293.         if (allocation[0][i]) {

  1294.             scale_factors[0][i] = get_bits(&s->gb, 6);

  1295.             scale_factors[1][i] = get_bits(&s->gb, 6);

  1296.         }

  1297.     }

  1298.     

  1299.     /* compute samples */

  1300.     for(j=0;j<12;j++) {

  1301.         for(i=0;i<bound;i++) {

  1302.             for(ch=0;ch<s->nb_channels;ch++) {

  1303.                 n = allocation[ch][i];

  1304.                 if (n) {

  1305.                     mant = get_bits(&s->gb, n + 1);

  1306.                     v = l1_unscale(n, mant, scale_factors[ch][i]);

  1307.                 } else {

  1308.                     v = 0;

  1309.                 }

  1310.                 s->sb_samples[ch][j][i] = v;

  1311.             }

  1312.         }

  1313.         for(i=bound;i<SBLIMIT;i++) {

  1314.             n = allocation[0][i];

  1315.             if (n) {

  1316.                 mant = get_bits(&s->gb, n + 1);

  1317.                 v = l1_unscale(n, mant, scale_factors[0][i]);

  1318.                 s->sb_samples[0][j][i] = v;

  1319.                 v = l1_unscale(n, mant, scale_factors[1][i]);

  1320.                 s->sb_samples[1][j][i] = v;

  1321.             } else {

  1322.                 s->sb_samples[0][j][i] = 0;

  1323.                 s->sb_samples[1][j][i] = 0;

  1324.             }

  1325.         }

  1326.     }

  1327.     return 12;

  1328. }

  1329. 

  1330. /* bitrate is in kb/s */

  1331. int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)

  1332. {

  1333.     int ch_bitrate, table;

  1334.     

  1335.     ch_bitrate = bitrate / nb_channels;

  1336.     if (!lsf) {

  1337.         if ((freq == 48000 && ch_bitrate >= 56) ||

  1338.             (ch_bitrate >= 56 && ch_bitrate <= 80)) 

  1339.             table = 0;

  1340.         else if (freq != 48000 && ch_bitrate >= 96) 

  1341.             table = 1;

  1342.         else if (freq != 32000 && ch_bitrate <= 48) 

  1343.             table = 2;

  1344.         else 

  1345.             table = 3;

  1346.     } else {

  1347.         table = 4;

  1348.     }

  1349.     return table;

  1350. }

  1351. 

  1352. static int mp_decode_layer2(MPADecodeContext *s)

  1353. {

  1354.     int sblimit; /* number of used subbands */

  1355.     const unsigned char *alloc_table;

  1356.     int table, bit_alloc_bits, i, j, ch, bound, v;

  1357.     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];

  1358.     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];

  1359.     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;

  1360.     int scale, qindex, bits, steps, k, l, m, b;

  1361. 

  1362.     /* select decoding table */

  1363.     table = l2_select_table(s->bit_rate / 1000, s->nb_channels, 

  1364.                             s->sample_rate, s->lsf);

  1365.     sblimit = sblimit_table[table];

  1366.     alloc_table = alloc_tables[table];

  1367. 

  1368.     if (s->mode == MPA_JSTEREO) 

  1369.         bound = (s->mode_ext + 1) * 4;

  1370.     else

  1371.         bound = sblimit;

  1372. 

  1373.     dprintf("bound=%d sblimit=%d\n", bound, sblimit);

  1374.     /* parse bit allocation */

  1375.     j = 0;

  1376.     for(i=0;i<bound;i++) {

  1377.         bit_alloc_bits = alloc_table[j];

  1378.         for(ch=0;ch<s->nb_channels;ch++) {

  1379.             bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);

  1380.         }

  1381.         j += 1 << bit_alloc_bits;

  1382.     }

  1383.     for(i=bound;i<sblimit;i++) {

  1384.         bit_alloc_bits = alloc_table[j];

  1385.         v = get_bits(&s->gb, bit_alloc_bits);

  1386.         bit_alloc[0][i] = v;

  1387.         bit_alloc[1][i] = v;

  1388.         j += 1 << bit_alloc_bits;

  1389.     }

  1390. 

  1391. #ifdef DEBUG

  1392.     {

  1393.         for(ch=0;ch<s->nb_channels;ch++) {

  1394.             for(i=0;i<sblimit;i++)

  1395.                 printf(" %d", bit_alloc[ch][i]);

  1396.             printf("\n");

  1397.         }

  1398.     }

  1399. #endif

  1400. 

  1401.     /* scale codes */

  1402.     for(i=0;i<sblimit;i++) {

  1403.         for(ch=0;ch<s->nb_channels;ch++) {

  1404.             if (bit_alloc[ch][i]) 

  1405.                 scale_code[ch][i] = get_bits(&s->gb, 2);

  1406.         }

  1407.     }

  1408.     

  1409.     /* scale factors */

  1410.     for(i=0;i<sblimit;i++) {

  1411.         for(ch=0;ch<s->nb_channels;ch++) {

  1412.             if (bit_alloc[ch][i]) {

  1413.                 sf = scale_factors[ch][i];

  1414.                 switch(scale_code[ch][i]) {

  1415.                 default:

  1416.                 case 0:

  1417.                     sf[0] = get_bits(&s->gb, 6);

  1418.                     sf[1] = get_bits(&s->gb, 6);

  1419.                     sf[2] = get_bits(&s->gb, 6);

  1420.                     break;

  1421.                 case 2:

  1422.                     sf[0] = get_bits(&s->gb, 6);

  1423.                     sf[1] = sf[0];

  1424.                     sf[2] = sf[0];

  1425.                     break;

  1426.                 case 1:

  1427.                     sf[0] = get_bits(&s->gb, 6);

  1428.                     sf[2] = get_bits(&s->gb, 6);

  1429.                     sf[1] = sf[0];

  1430.                     break;

  1431.                 case 3:

  1432.                     sf[0] = get_bits(&s->gb, 6);

  1433.                     sf[2] = get_bits(&s->gb, 6);

  1434.                     sf[1] = sf[2];

  1435.                     break;

  1436.                 }

  1437.             }

  1438.         }

  1439.     }

  1440. 

  1441. #ifdef DEBUG

  1442.     for(ch=0;ch<s->nb_channels;ch++) {

  1443.         for(i=0;i<sblimit;i++) {

  1444.             if (bit_alloc[ch][i]) {

  1445.                 sf = scale_factors[ch][i];

  1446.                 printf(" %d %d %d", sf[0], sf[1], sf[2]);

  1447.             } else {

  1448.                 printf(" -");

  1449.             }

  1450.         }

  1451.         printf("\n");

  1452.     }

  1453. #endif

  1454. 

  1455.     /* samples */

  1456.     for(k=0;k<3;k++) {

  1457.         for(l=0;l<12;l+=3) {

  1458.             j = 0;

  1459.             for(i=0;i<bound;i++) {

  1460.                 bit_alloc_bits = alloc_table[j];

  1461.                 for(ch=0;ch<s->nb_channels;ch++) {

  1462.                     b = bit_alloc[ch][i];

  1463.                     if (b) {

  1464.                         scale = scale_factors[ch][i][k];

  1465.                         qindex = alloc_table[j+b];

  1466.                         bits = quant_bits[qindex];

  1467.                         if (bits < 0) {

  1468.                             /* 3 values at the same time */

  1469.                             v = get_bits(&s->gb, -bits);

  1470.                             steps = quant_steps[qindex];

  1471.                             s->sb_samples[ch][k * 12 + l + 0][i] = 

  1472.                                 l2_unscale_group(steps, v % steps, scale);

  1473.                             v = v / steps;

  1474.                             s->sb_samples[ch][k * 12 + l + 1][i] = 

  1475.                                 l2_unscale_group(steps, v % steps, scale);

  1476.                             v = v / steps;

  1477.                             s->sb_samples[ch][k * 12 + l + 2][i] = 

  1478.                                 l2_unscale_group(steps, v, scale);

  1479.                         } else {

  1480.                             for(m=0;m<3;m++) {

  1481.                                 v = get_bits(&s->gb, bits);

  1482.                                 v = l1_unscale(bits - 1, v, scale);

  1483.                                 s->sb_samples[ch][k * 12 + l + m][i] = v;

  1484.                             }

  1485.                         }

  1486.                     } else {

  1487.                         s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1488.                         s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1489.                         s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1490.                     }

  1491.                 }

  1492.                 /* next subband in alloc table */

  1493.                 j += 1 << bit_alloc_bits; 

  1494.             }

  1495.             /* XXX: find a way to avoid this duplication of code */

  1496.             for(i=bound;i<sblimit;i++) {

  1497.                 bit_alloc_bits = alloc_table[j];

  1498.                 b = bit_alloc[0][i];

  1499.                 if (b) {

  1500.                     int mant, scale0, scale1;

  1501.                     scale0 = scale_factors[0][i][k];

  1502.                     scale1 = scale_factors[1][i][k];

  1503.                     qindex = alloc_table[j+b];

  1504.                     bits = quant_bits[qindex];

  1505.                     if (bits < 0) {

  1506.                         /* 3 values at the same time */

  1507.                         v = get_bits(&s->gb, -bits);

  1508.                         steps = quant_steps[qindex];

  1509.                         mant = v % steps;

  1510.                         v = v / steps;

  1511.                         s->sb_samples[0][k * 12 + l + 0][i] = 

  1512.                             l2_unscale_group(steps, mant, scale0);

  1513.                         s->sb_samples[1][k * 12 + l + 0][i] = 

  1514.                             l2_unscale_group(steps, mant, scale1);

  1515.                         mant = v % steps;

  1516.                         v = v / steps;

  1517.                         s->sb_samples[0][k * 12 + l + 1][i] = 

  1518.                             l2_unscale_group(steps, mant, scale0);

  1519.                         s->sb_samples[1][k * 12 + l + 1][i] = 

  1520.                             l2_unscale_group(steps, mant, scale1);

  1521.                         s->sb_samples[0][k * 12 + l + 2][i] = 

  1522.                             l2_unscale_group(steps, v, scale0);

  1523.                         s->sb_samples[1][k * 12 + l + 2][i] = 

  1524.                             l2_unscale_group(steps, v, scale1);

  1525.                     } else {

  1526.                         for(m=0;m<3;m++) {

  1527.                             mant = get_bits(&s->gb, bits);

  1528.                             s->sb_samples[0][k * 12 + l + m][i] = 

  1529.                                 l1_unscale(bits - 1, mant, scale0);

  1530.                             s->sb_samples[1][k * 12 + l + m][i] = 

  1531.                                 l1_unscale(bits - 1, mant, scale1);

  1532.                         }

  1533.                     }

  1534.                 } else {

  1535.                     s->sb_samples[0][k * 12 + l + 0][i] = 0;

  1536.                     s->sb_samples[0][k * 12 + l + 1][i] = 0;

  1537.                     s->sb_samples[0][k * 12 + l + 2][i] = 0;

  1538.                     s->sb_samples[1][k * 12 + l + 0][i] = 0;

  1539.                     s->sb_samples[1][k * 12 + l + 1][i] = 0;

  1540.                     s->sb_samples[1][k * 12 + l + 2][i] = 0;

  1541.                 }

  1542.                 /* next subband in alloc table */

  1543.                 j += 1 << bit_alloc_bits; 

  1544.             }

  1545.             /* fill remaining samples to zero */

  1546.             for(i=sblimit;i<SBLIMIT;i++) {

  1547.                 for(ch=0;ch<s->nb_channels;ch++) {

  1548.                     s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1549.                     s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1550.                     s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1551.                 }

  1552.             }

  1553.         }

  1554.     }

  1555.     return 3 * 12;

  1556. }

  1557. 

  1558. /*

  1559.  * Seek back in the stream for backstep bytes (at most 511 bytes)

  1560.  */

  1561. static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)

  1562. {

  1563.     uint8_t *ptr;

  1564. 

  1565.     /* compute current position in stream */

  1566.     ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));

  1567. 

  1568.     /* copy old data before current one */

  1569.     ptr -= backstep;

  1570.     memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] + 

  1571.            BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);

  1572.     /* init get bits again */

  1573.     init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);

  1574. 

  1575.     /* prepare next buffer */

  1576.     s->inbuf_index ^= 1;

  1577.     s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];

  1578.     s->old_frame_size = s->frame_size;

  1579. }

  1580. 

  1581. static inline void lsf_sf_expand(int *slen,

  1582.                                  int sf, int n1, int n2, int n3)

  1583. {

  1584.     if (n3) {

  1585.         slen[3] = sf % n3;

  1586.         sf /= n3;

  1587.     } else {

  1588.         slen[3] = 0;

  1589.     }

  1590.     if (n2) {

  1591.         slen[2] = sf % n2;

  1592.         sf /= n2;

  1593.     } else {

  1594.         slen[2] = 0;

  1595.     }

  1596.     slen[1] = sf % n1;

  1597.     sf /= n1;

  1598.     slen[0] = sf;

  1599. }

  1600. 

  1601. static void exponents_from_scale_factors(MPADecodeContext *s, 

  1602.                                          GranuleDef *g,

  1603.                                          int16_t *exponents)

  1604. {

  1605.     const uint8_t *bstab, *pretab;

  1606.     int len, i, j, k, l, v0, shift, gain, gains[3];

  1607.     int16_t *exp_ptr;

  1608. 

  1609.     exp_ptr = exponents;

  1610.     gain = g->global_gain - 210;

  1611.     shift = g->scalefac_scale + 1;

  1612. 

  1613.     bstab = band_size_long[s->sample_rate_index];

  1614.     pretab = mpa_pretab[g->preflag];

  1615.     for(i=0;i<g->long_end;i++) {

  1616.         v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);

  1617.         len = bstab[i];

  1618.         for(j=len;j>0;j--)

  1619.             *exp_ptr++ = v0;

  1620.     }

  1621. 

  1622.     if (g->short_start < 13) {

  1623.         bstab = band_size_short[s->sample_rate_index];

  1624.         gains[0] = gain - (g->subblock_gain[0] << 3);

  1625.         gains[1] = gain - (g->subblock_gain[1] << 3);

  1626.         gains[2] = gain - (g->subblock_gain[2] << 3);

  1627.         k = g->long_end;

  1628.         for(i=g->short_start;i<13;i++) {

  1629.             len = bstab[i];

  1630.             for(l=0;l<3;l++) {

  1631.                 v0 = gains[l] - (g->scale_factors[k++] << shift);

  1632.                 for(j=len;j>0;j--)

  1633.                 *exp_ptr++ = v0;

  1634.             }

  1635.         }

  1636.     }

  1637. }

  1638. 

  1639. /* handle n = 0 too */

  1640. static inline int get_bitsz(GetBitContext *s, int n)

  1641. {

  1642.     if (n == 0)

  1643.         return 0;

  1644.     else

  1645.         return get_bits(s, n);

  1646. }

  1647. 

  1648. static int huffman_decode(MPADecodeContext *s, GranuleDef *g,

  1649.                           int16_t *exponents, int end_pos)

  1650. {

  1651.     int s_index;

  1652.     int linbits, code, x, y, l, v, i, j, k, pos;

  1653.     GetBitContext last_gb;

  1654.     VLC *vlc;

  1655.     uint8_t *code_table;

  1656. 

  1657.     /* low frequencies (called big values) */

  1658.     s_index = 0;

  1659.     for(i=0;i<3;i++) {

  1660.         j = g->region_size[i];

  1661.         if (j == 0)

  1662.             continue;

  1663.         /* select vlc table */

  1664.         k = g->table_select[i];

  1665.         l = mpa_huff_data[k][0];

  1666.         linbits = mpa_huff_data[k][1];

  1667.         vlc = &huff_vlc[l];

  1668.         code_table = huff_code_table[l];

  1669. 

  1670.         /* read huffcode and compute each couple */

  1671.         for(;j>0;j--) {

  1672.             if (get_bits_count(&s->gb) >= end_pos)

  1673.                 break;

  1674.             if (code_table) {

  1675.                 code = get_vlc(&s->gb, vlc);

  1676.                 if (code < 0)

  1677.                     return -1;

  1678.                 y = code_table[code];

  1679.                 x = y >> 4;

  1680.                 y = y & 0x0f;

  1681.             } else {

  1682.                 x = 0;

  1683.                 y = 0;

  1684.             }

  1685.             dprintf("region=%d n=%d x=%d y=%d exp=%d\n", 

  1686.                     i, g->region_size[i] - j, x, y, exponents[s_index]);

  1687.             if (x) {

  1688.                 if (x == 15)

  1689.                     x += get_bitsz(&s->gb, linbits);

  1690.                 v = l3_unscale(x, exponents[s_index]);

  1691.                 if (get_bits1(&s->gb))

  1692.                     v = -v;

  1693.             } else {

  1694.                 v = 0;

  1695.             }

  1696.             g->sb_hybrid[s_index++] = v;

  1697.             if (y) {

  1698.                 if (y == 15)

  1699.                     y += get_bitsz(&s->gb, linbits);

  1700.                 v = l3_unscale(y, exponents[s_index]);

  1701.                 if (get_bits1(&s->gb))

  1702.                     v = -v;

  1703.             } else {

  1704.                 v = 0;

  1705.             }

  1706.             g->sb_hybrid[s_index++] = v;

  1707.         }

  1708.     }

  1709.             

  1710.     /* high frequencies */

  1711.     vlc = &huff_quad_vlc[g->count1table_select];

  1712.     last_gb.buffer = NULL;

  1713.     while (s_index <= 572) {

  1714.         pos = get_bits_count(&s->gb);

  1715.         if (pos >= end_pos) {

  1716.             if (pos > end_pos && last_gb.buffer != NULL) {

  1717.                 /* some encoders generate an incorrect size for this

  1718.                    part. We must go back into the data */

  1719.                 s_index -= 4;

  1720.                 s->gb = last_gb;

  1721.             }

  1722.             break;

  1723.         }

  1724.         last_gb= s->gb;

  1725. 

  1726.         code = get_vlc(&s->gb, vlc);

  1727.         dprintf("t=%d code=%d\n", g->count1table_select, code);

  1728.         if (code < 0)

  1729.             return -1;

  1730.         for(i=0;i<4;i++) {

  1731.             if (code & (8 >> i)) {

  1732.                 /* non zero value. Could use a hand coded function for

  1733.                    'one' value */

  1734.                 v = l3_unscale(1, exponents[s_index]);

  1735.                 if(get_bits1(&s->gb))

  1736.                     v = -v;

  1737.             } else {

  1738.                 v = 0;

  1739.             }

  1740.             g->sb_hybrid[s_index++] = v;

  1741.         }

  1742.     }

  1743.     while (s_index < 576)

  1744.         g->sb_hybrid[s_index++] = 0;

  1745.     return 0;

  1746. }

  1747. 

  1748. /* Reorder short blocks from bitstream order to interleaved order. It

  1749.    would be faster to do it in parsing, but the code would be far more

  1750.    complicated */

  1751. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1752. {

  1753.     int i, j, k, len;

  1754.     int32_t *ptr, *dst, *ptr1;

  1755.     int32_t tmp[576];

  1756. 

  1757.     if (g->block_type != 2)

  1758.         return;

  1759. 

  1760.     if (g->switch_point) {

  1761.         if (s->sample_rate_index != 8) {

  1762.             ptr = g->sb_hybrid + 36;

  1763.         } else {

  1764.             ptr = g->sb_hybrid + 48;

  1765.         }

  1766.     } else {

  1767.         ptr = g->sb_hybrid;

  1768.     }

  1769.     

  1770.     for(i=g->short_start;i<13;i++) {

  1771.         len = band_size_short[s->sample_rate_index][i];

  1772.         ptr1 = ptr;

  1773.         for(k=0;k<3;k++) {

  1774.             dst = tmp + k;

  1775.             for(j=len;j>0;j--) {

  1776.                 *dst = *ptr++;

  1777.                 dst += 3;

  1778.             }

  1779.         }

  1780.         memcpy(ptr1, tmp, len * 3 * sizeof(int32_t));

  1781.     }

  1782. }

  1783. 

  1784. #define ISQRT2 FIXR(0.70710678118654752440)

  1785. 

  1786. static void compute_stereo(MPADecodeContext *s,

  1787.                            GranuleDef *g0, GranuleDef *g1)

  1788. {

  1789.     int i, j, k, l;

  1790.     int32_t v1, v2;

  1791.     int sf_max, tmp0, tmp1, sf, len, non_zero_found;

  1792.     int32_t (*is_tab)[16];

  1793.     int32_t *tab0, *tab1;

  1794.     int non_zero_found_short[3];

  1795. 

  1796.     /* intensity stereo */

  1797.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1798.         if (!s->lsf) {

  1799.             is_tab = is_table;

  1800.             sf_max = 7;

  1801.         } else {

  1802.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1803.             sf_max = 16;

  1804.         }

  1805.             

  1806.         tab0 = g0->sb_hybrid + 576;

  1807.         tab1 = g1->sb_hybrid + 576;

  1808. 

  1809.         non_zero_found_short[0] = 0;

  1810.         non_zero_found_short[1] = 0;

  1811.         non_zero_found_short[2] = 0;

  1812.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1813.         for(i = 12;i >= g1->short_start;i--) {

  1814.             /* for last band, use previous scale factor */

  1815.             if (i != 11)

  1816.                 k -= 3;

  1817.             len = band_size_short[s->sample_rate_index][i];

  1818.             for(l=2;l>=0;l--) {

  1819.                 tab0 -= len;

  1820.                 tab1 -= len;

  1821.                 if (!non_zero_found_short[l]) {

  1822.                     /* test if non zero band. if so, stop doing i-stereo */

  1823.                     for(j=0;j<len;j++) {

  1824.                         if (tab1[j] != 0) {

  1825.                             non_zero_found_short[l] = 1;

  1826.                             goto found1;

  1827.                         }

  1828.                     }

  1829.                     sf = g1->scale_factors[k + l];

  1830.                     if (sf >= sf_max)

  1831.                         goto found1;

  1832. 

  1833.                     v1 = is_tab[0][sf];

  1834.                     v2 = is_tab[1][sf];

  1835.                     for(j=0;j<len;j++) {

  1836.                         tmp0 = tab0[j];

  1837.                         tab0[j] = MULL(tmp0, v1);

  1838.                         tab1[j] = MULL(tmp0, v2);

  1839.                     }

  1840.                 } else {

  1841.                 found1:

  1842.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1843.                         /* lower part of the spectrum : do ms stereo

  1844.                            if enabled */

  1845.                         for(j=0;j<len;j++) {

  1846.                             tmp0 = tab0[j];

  1847.                             tmp1 = tab1[j];

  1848.                             tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1849.                             tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1850.                         }

  1851.                     }

  1852.                 }

  1853.             }

  1854.         }

  1855. 

  1856.         non_zero_found = non_zero_found_short[0] | 

  1857.             non_zero_found_short[1] | 

  1858.             non_zero_found_short[2];

  1859. 

  1860.         for(i = g1->long_end - 1;i >= 0;i--) {

  1861.             len = band_size_long[s->sample_rate_index][i];

  1862.             tab0 -= len;

  1863.             tab1 -= len;

  1864.             /* test if non zero band. if so, stop doing i-stereo */

  1865.             if (!non_zero_found) {

  1866.                 for(j=0;j<len;j++) {

  1867.                     if (tab1[j] != 0) {

  1868.                         non_zero_found = 1;

  1869.                         goto found2;

  1870.                     }

  1871.                 }

  1872.                 /* for last band, use previous scale factor */

  1873.                 k = (i == 21) ? 20 : i;

  1874.                 sf = g1->scale_factors[k];

  1875.                 if (sf >= sf_max)

  1876.                     goto found2;

  1877.                 v1 = is_tab[0][sf];

  1878.                 v2 = is_tab[1][sf];

  1879.                 for(j=0;j<len;j++) {

  1880.                     tmp0 = tab0[j];

  1881.                     tab0[j] = MULL(tmp0, v1);

  1882.                     tab1[j] = MULL(tmp0, v2);

  1883.                 }

  1884.             } else {

  1885.             found2:

  1886.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1887.                     /* lower part of the spectrum : do ms stereo

  1888.                        if enabled */

  1889.                     for(j=0;j<len;j++) {

  1890.                         tmp0 = tab0[j];

  1891.                         tmp1 = tab1[j];

  1892.                         tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1893.                         tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1894.                     }

  1895.                 }

  1896.             }

  1897.         }

  1898.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1899.         /* ms stereo ONLY */

  1900.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1901.            global gain */

  1902.         tab0 = g0->sb_hybrid;

  1903.         tab1 = g1->sb_hybrid;

  1904.         for(i=0;i<576;i++) {

  1905.             tmp0 = tab0[i];

  1906.             tmp1 = tab1[i];

  1907.             tab0[i] = tmp0 + tmp1;

  1908.             tab1[i] = tmp0 - tmp1;

  1909.         }

  1910.     }

  1911. }

  1912. 

  1913. static void compute_antialias_integer(MPADecodeContext *s,

  1914.                               GranuleDef *g)

  1915. {

  1916.     int32_t *ptr, *p0, *p1, *csa;

  1917.     int n, i, j;

  1918. 

  1919.     /* we antialias only "long" bands */

  1920.     if (g->block_type == 2) {

  1921.         if (!g->switch_point)

  1922.             return;

  1923.         /* XXX: check this for 8000Hz case */

  1924.         n = 1;

  1925.     } else {

  1926.         n = SBLIMIT - 1;

  1927.     }

  1928.     

  1929.     ptr = g->sb_hybrid + 18;

  1930.     for(i = n;i > 0;i--) {

  1931.         p0 = ptr - 1;

  1932.         p1 = ptr;

  1933.         csa = &csa_table[0][0];       

  1934.         for(j=0;j<4;j++) {

  1935.             int tmp0 = *p0;

  1936.             int tmp1 = *p1;

  1937. #if 0

  1938.             *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));

  1939.             *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));

  1940. #else

  1941.             int64_t tmp2= MUL64(tmp0 + tmp1, csa[0]);

  1942.             *p0 = FRAC_RND(tmp2 - MUL64(tmp1, csa[2]));

  1943.             *p1 = FRAC_RND(tmp2 + MUL64(tmp0, csa[3]));

  1944. #endif

  1945.             p0--; p1++;

  1946.             csa += 4;

  1947.             tmp0 = *p0;

  1948.             tmp1 = *p1;

  1949. #if 0

  1950.             *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));

  1951.             *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));

  1952. #else

  1953.             tmp2= MUL64(tmp0 + tmp1, csa[0]);

  1954.             *p0 = FRAC_RND(tmp2 - MUL64(tmp1, csa[2]));

  1955.             *p1 = FRAC_RND(tmp2 + MUL64(tmp0, csa[3]));

  1956. #endif

  1957.             p0--; p1++;

  1958.             csa += 4;

  1959.         }

  1960.         ptr += 18;       

  1961.     }

  1962. }

  1963. 

  1964. static void compute_antialias_float(MPADecodeContext *s,

  1965.                               GranuleDef *g)

  1966. {

  1967.     int32_t *ptr, *p0, *p1;

  1968.     int n, i, j;

  1969. 

  1970.     /* we antialias only "long" bands */

  1971.     if (g->block_type == 2) {

  1972.         if (!g->switch_point)

  1973.             return;

  1974.         /* XXX: check this for 8000Hz case */

  1975.         n = 1;

  1976.     } else {

  1977.         n = SBLIMIT - 1;

  1978.     }

  1979.     

  1980.     ptr = g->sb_hybrid + 18;

  1981.     for(i = n;i > 0;i--) {

  1982.         float *csa = &csa_table_float[0][0];       

  1983.         p0 = ptr - 1;

  1984.         p1 = ptr;

  1985.         for(j=0;j<4;j++) {

  1986.             float tmp0 = *p0;

  1987.             float tmp1 = *p1;

  1988. #if 1

  1989.             *p0 = lrintf(tmp0 * csa[0] - tmp1 * csa[1]);

  1990.             *p1 = lrintf(tmp0 * csa[1] + tmp1 * csa[0]);

  1991. #else

  1992.             float tmp2= (tmp0 + tmp1) * csa[0];

  1993.             *p0 = lrintf(tmp2 - tmp1 * csa[2]);

  1994.             *p1 = lrintf(tmp2 + tmp0 * csa[3]);

  1995. #endif

  1996.             p0--; p1++;

  1997.             csa += 4;

  1998.             tmp0 = *p0;

  1999.             tmp1 = *p1;

  2000. #if 1

  2001.             *p0 = lrintf(tmp0 * csa[0] - tmp1 * csa[1]);

  2002.             *p1 = lrintf(tmp0 * csa[1] + tmp1 * csa[0]);

  2003. #else

  2004.             tmp2= (tmp0 + tmp1) * csa[0];

  2005.             *p0 = lrintf(tmp2 - tmp1 * csa[2]);

  2006.             *p1 = lrintf(tmp2 + tmp0 * csa[3]);

  2007. #endif

  2008.             p0--; p1++;

  2009.             csa += 4;

  2010.         }

  2011.         ptr += 18;       

  2012.     }

  2013. }

  2014. 

  2015. static void compute_imdct(MPADecodeContext *s,

  2016.                           GranuleDef *g, 

  2017.                           int32_t *sb_samples,

  2018.                           int32_t *mdct_buf)

  2019. {

  2020.     int32_t *ptr, *win, *win1, *buf, *buf2, *out_ptr, *ptr1;

  2021.     int32_t in[6];

  2022.     int32_t out[36];

  2023.     int32_t out2[12];

  2024.     int i, j, k, mdct_long_end, v, sblimit;

  2025. 

  2026.     /* find last non zero block */

  2027.     ptr = g->sb_hybrid + 576;

  2028.     ptr1 = g->sb_hybrid + 2 * 18;

  2029.     while (ptr >= ptr1) {

  2030.         ptr -= 6;

  2031.         v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];

  2032.         if (v != 0)

  2033.             break;

  2034.     }

  2035.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  2036. 

  2037.     if (g->block_type == 2) {

  2038.         /* XXX: check for 8000 Hz */

  2039.         if (g->switch_point)

  2040.             mdct_long_end = 2;

  2041.         else

  2042.             mdct_long_end = 0;

  2043.     } else {

  2044.         mdct_long_end = sblimit;

  2045.     }

  2046. 

  2047.     buf = mdct_buf;

  2048.     ptr = g->sb_hybrid;

  2049.     for(j=0;j<mdct_long_end;j++) {

  2050.         imdct36(out, ptr);

  2051.         /* apply window & overlap with previous buffer */

  2052.         out_ptr = sb_samples + j;

  2053.         /* select window */

  2054.         if (g->switch_point && j < 2)

  2055.             win1 = mdct_win[0];

  2056.         else

  2057.             win1 = mdct_win[g->block_type];

  2058.         /* select frequency inversion */

  2059.         win = win1 + ((4 * 36) & -(j & 1));

  2060.         for(i=0;i<18;i++) {

  2061.             *out_ptr = MULL(out[i], win[i]) + buf[i];

  2062.             buf[i] = MULL(out[i + 18], win[i + 18]);

  2063.             out_ptr += SBLIMIT;

  2064.         }

  2065.         ptr += 18;

  2066.         buf += 18;

  2067.     }

  2068.     for(j=mdct_long_end;j<sblimit;j++) {

  2069.         for(i=0;i<6;i++) {

  2070.             out[i] = 0;

  2071.             out[6 + i] = 0;

  2072.             out[30+i] = 0;

  2073.         }

  2074.         /* select frequency inversion */

  2075.         win = mdct_win[2] + ((4 * 36) & -(j & 1));

  2076.         buf2 = out + 6;

  2077.         for(k=0;k<3;k++) {

  2078.             /* reorder input for short mdct */

  2079.             ptr1 = ptr + k;

  2080.             for(i=0;i<6;i++) {

  2081.                 in[i] = *ptr1;

  2082.                 ptr1 += 3;

  2083.             }

  2084.             imdct12(out2, in);

  2085.             /* apply 12 point window and do small overlap */

  2086.             for(i=0;i<6;i++) {

  2087.                 buf2[i] = MULL(out2[i], win[i]) + buf2[i];

  2088.                 buf2[i + 6] = MULL(out2[i + 6], win[i + 6]);

  2089.             }

  2090.             buf2 += 6;

  2091.         }

  2092.         /* overlap */

  2093.         out_ptr = sb_samples + j;

  2094.         for(i=0;i<18;i++) {

  2095.             *out_ptr = out[i] + buf[i];

  2096.             buf[i] = out[i + 18];

  2097.             out_ptr += SBLIMIT;

  2098.         }

  2099.         ptr += 18;

  2100.         buf += 18;

  2101.     }

  2102.     /* zero bands */

  2103.     for(j=sblimit;j<SBLIMIT;j++) {

  2104.         /* overlap */

  2105.         out_ptr = sb_samples + j;

  2106.         for(i=0;i<18;i++) {

  2107.             *out_ptr = buf[i];

  2108.             buf[i] = 0;

  2109.             out_ptr += SBLIMIT;

  2110.         }

  2111.         buf += 18;

  2112.     }

  2113. }

  2114. 

  2115. #if defined(DEBUG)

  2116. void sample_dump(int fnum, int32_t *tab, int n)

  2117. {

  2118.     static FILE *files[16], *f;

  2119.     char buf[512];

  2120.     int i;

  2121.     int32_t v;

  2122.     

  2123.     f = files[fnum];

  2124.     if (!f) {

  2125.         sprintf(buf, "/tmp/out%d.%s.pcm", 

  2126.                 fnum, 

  2127. #ifdef USE_HIGHPRECISION

  2128.                 "hp"

  2129. #else

  2130.                 "lp"

  2131. #endif

  2132.                 );

  2133.         f = fopen(buf, "w");

  2134.         if (!f)

  2135.             return;

  2136.         files[fnum] = f;

  2137.     }

  2138.     

  2139.     if (fnum == 0) {

  2140.         static int pos = 0;

  2141.         printf("pos=%d\n", pos);

  2142.         for(i=0;i<n;i++) {

  2143.             printf(" %0.4f", (double)tab[i] / FRAC_ONE);

  2144.             if ((i % 18) == 17)

  2145.                 printf("\n");

  2146.         }

  2147.         pos += n;

  2148.     }

  2149.     for(i=0;i<n;i++) {

  2150.         /* normalize to 23 frac bits */

  2151.         v = tab[i] << (23 - FRAC_BITS);

  2152.         fwrite(&v, 1, sizeof(int32_t), f);

  2153.     }

  2154. }

  2155. #endif

  2156. 

  2157. 

  2158. /* main layer3 decoding function */

  2159. static int mp_decode_layer3(MPADecodeContext *s)

  2160. {

  2161.     int nb_granules, main_data_begin, private_bits;

  2162.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;

  2163.     GranuleDef granules[2][2], *g;

  2164.     int16_t exponents[576];

  2165. 

  2166.     /* read side info */

  2167.     if (s->lsf) {

  2168.         main_data_begin = get_bits(&s->gb, 8);

  2169.         if (s->nb_channels == 2)

  2170.             private_bits = get_bits(&s->gb, 2);

  2171.         else

  2172.             private_bits = get_bits(&s->gb, 1);

  2173.         nb_granules = 1;

  2174.     } else {

  2175.         main_data_begin = get_bits(&s->gb, 9);

  2176.         if (s->nb_channels == 2)

  2177.             private_bits = get_bits(&s->gb, 3);

  2178.         else

  2179.             private_bits = get_bits(&s->gb, 5);

  2180.         nb_granules = 2;

  2181.         for(ch=0;ch<s->nb_channels;ch++) {

  2182.             granules[ch][0].scfsi = 0; /* all scale factors are transmitted */

  2183.             granules[ch][1].scfsi = get_bits(&s->gb, 4);

  2184.         }

  2185.     }

  2186.     

  2187.     for(gr=0;gr<nb_granules;gr++) {

  2188.         for(ch=0;ch<s->nb_channels;ch++) {

  2189.             dprintf("gr=%d ch=%d: side_info\n", gr, ch);

  2190.             g = &granules[ch][gr];

  2191.             g->part2_3_length = get_bits(&s->gb, 12);

  2192.             g->big_values = get_bits(&s->gb, 9);

  2193.             g->global_gain = get_bits(&s->gb, 8);

  2194.             /* if MS stereo only is selected, we precompute the

  2195.                1/sqrt(2) renormalization factor */

  2196.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) == 

  2197.                 MODE_EXT_MS_STEREO)

  2198.                 g->global_gain -= 2;

  2199.             if (s->lsf)

  2200.                 g->scalefac_compress = get_bits(&s->gb, 9);

  2201.             else

  2202.                 g->scalefac_compress = get_bits(&s->gb, 4);

  2203.             blocksplit_flag = get_bits(&s->gb, 1);

  2204.             if (blocksplit_flag) {

  2205.                 g->block_type = get_bits(&s->gb, 2);

  2206.                 if (g->block_type == 0)

  2207.                     return -1;

  2208.                 g->switch_point = get_bits(&s->gb, 1);

  2209.                 for(i=0;i<2;i++)

  2210.                     g->table_select[i] = get_bits(&s->gb, 5);

  2211.                 for(i=0;i<3;i++) 

  2212.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  2213.                 /* compute huffman coded region sizes */

  2214.                 if (g->block_type == 2)

  2215.                     g->region_size[0] = (36 / 2);

  2216.                 else {

  2217.                     if (s->sample_rate_index <= 2) 

  2218.                         g->region_size[0] = (36 / 2);

  2219.                     else if (s->sample_rate_index != 8) 

  2220.                         g->region_size[0] = (54 / 2);

  2221.                     else

  2222.                         g->region_size[0] = (108 / 2);

  2223.                 }

  2224.                 g->region_size[1] = (576 / 2);

  2225.             } else {

  2226.                 int region_address1, region_address2, l;

  2227.                 g->block_type = 0;

  2228.                 g->switch_point = 0;

  2229.                 for(i=0;i<3;i++)

  2230.                     g->table_select[i] = get_bits(&s->gb, 5);

  2231.                 /* compute huffman coded region sizes */

  2232.                 region_address1 = get_bits(&s->gb, 4);

  2233.                 region_address2 = get_bits(&s->gb, 3);

  2234.                 dprintf("region1=%d region2=%d\n", 

  2235.                         region_address1, region_address2);

  2236.                 g->region_size[0] = 

  2237.                     band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;

  2238.                 l = region_address1 + region_address2 + 2;

  2239.                 /* should not overflow */

  2240.                 if (l > 22)

  2241.                     l = 22;

  2242.                 g->region_size[1] = 

  2243.                     band_index_long[s->sample_rate_index][l] >> 1;

  2244.             }

  2245.             /* convert region offsets to region sizes and truncate

  2246.                size to big_values */

  2247.             g->region_size[2] = (576 / 2);

  2248.             j = 0;

  2249.             for(i=0;i<3;i++) {

  2250.                 k = g->region_size[i];

  2251.                 if (k > g->big_values)

  2252.                     k = g->big_values;

  2253.                 g->region_size[i] = k - j;

  2254.                 j = k;

  2255.             }

  2256. 

  2257.             /* compute band indexes */

  2258.             if (g->block_type == 2) {

  2259.                 if (g->switch_point) {

  2260.                     /* if switched mode, we handle the 36 first samples as

  2261.                        long blocks.  For 8000Hz, we handle the 48 first

  2262.                        exponents as long blocks (XXX: check this!) */

  2263.                     if (s->sample_rate_index <= 2)

  2264.                         g->long_end = 8;

  2265.                     else if (s->sample_rate_index != 8)

  2266.                         g->long_end = 6;

  2267.                     else

  2268.                         g->long_end = 4; /* 8000 Hz */

  2269.                     

  2270.                     if (s->sample_rate_index != 8)

  2271.                         g->short_start = 3;

  2272.                     else

  2273.                         g->short_start = 2; 

  2274.                 } else {

  2275.                     g->long_end = 0;

  2276.                     g->short_start = 0;

  2277.                 }

  2278.             } else {

  2279.                 g->short_start = 13;

  2280.                 g->long_end = 22;

  2281.             }

  2282.             

  2283.             g->preflag = 0;

  2284.             if (!s->lsf)

  2285.                 g->preflag = get_bits(&s->gb, 1);

  2286.             g->scalefac_scale = get_bits(&s->gb, 1);

  2287.             g->count1table_select = get_bits(&s->gb, 1);

  2288.             dprintf("block_type=%d switch_point=%d\n",

  2289.                     g->block_type, g->switch_point);

  2290.         }

  2291.     }

  2292. 

  2293.     /* now we get bits from the main_data_begin offset */

  2294.     dprintf("seekback: %d\n", main_data_begin);

  2295.     seek_to_maindata(s, main_data_begin);

  2296. 

  2297.     for(gr=0;gr<nb_granules;gr++) {

  2298.         for(ch=0;ch<s->nb_channels;ch++) {

  2299.             g = &granules[ch][gr];

  2300.             

  2301.             bits_pos = get_bits_count(&s->gb);

  2302.             

  2303.             if (!s->lsf) {

  2304.                 uint8_t *sc;

  2305.                 int slen, slen1, slen2;

  2306. 

  2307.                 /* MPEG1 scale factors */

  2308.                 slen1 = slen_table[0][g->scalefac_compress];

  2309.                 slen2 = slen_table[1][g->scalefac_compress];

  2310.                 dprintf("slen1=%d slen2=%d\n", slen1, slen2);

  2311.                 if (g->block_type == 2) {

  2312.                     n = g->switch_point ? 17 : 18;

  2313.                     j = 0;

  2314.                     for(i=0;i<n;i++)

  2315.                         g->scale_factors[j++] = get_bitsz(&s->gb, slen1);

  2316.                     for(i=0;i<18;i++)

  2317.                         g->scale_factors[j++] = get_bitsz(&s->gb, slen2);

  2318.                     for(i=0;i<3;i++)

  2319.                         g->scale_factors[j++] = 0;

  2320.                 } else {

  2321.                     sc = granules[ch][0].scale_factors;

  2322.                     j = 0;

  2323.                     for(k=0;k<4;k++) {

  2324.                         n = (k == 0 ? 6 : 5);

  2325.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  2326.                             slen = (k < 2) ? slen1 : slen2;

  2327.                             for(i=0;i<n;i++)

  2328.                                 g->scale_factors[j++] = get_bitsz(&s->gb, slen);

  2329.                         } else {

  2330.                             /* simply copy from last granule */

  2331.                             for(i=0;i<n;i++) {

  2332.                                 g->scale_factors[j] = sc[j];

  2333.                                 j++;

  2334.                             }

  2335.                         }

  2336.                     }

  2337.                     g->scale_factors[j++] = 0;

  2338.                 }

  2339. #if defined(DEBUG)

  2340.                 {

  2341.                     printf("scfsi=%x gr=%d ch=%d scale_factors:\n", 

  2342.                            g->scfsi, gr, ch);

  2343.                     for(i=0;i<j;i++)

  2344.                         printf(" %d", g->scale_factors[i]);

  2345.                     printf("\n");

  2346.                 }

  2347. #endif

  2348.             } else {

  2349.                 int tindex, tindex2, slen[4], sl, sf;

  2350. 

  2351.                 /* LSF scale factors */

  2352.                 if (g->block_type == 2) {

  2353.                     tindex = g->switch_point ? 2 : 1;

  2354.                 } else {

  2355.                     tindex = 0;

  2356.                 }

  2357.                 sf = g->scalefac_compress;

  2358.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  2359.                     /* intensity stereo case */

  2360.                     sf >>= 1;

  2361.                     if (sf < 180) {

  2362.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  2363.                         tindex2 = 3;

  2364.                     } else if (sf < 244) {

  2365.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  2366.                         tindex2 = 4;

  2367.                     } else {

  2368.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  2369.                         tindex2 = 5;

  2370.                     }

  2371.                 } else {

  2372.                     /* normal case */

  2373.                     if (sf < 400) {

  2374.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  2375.                         tindex2 = 0;

  2376.                     } else if (sf < 500) {

  2377.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  2378.                         tindex2 = 1;

  2379.                     } else {

  2380.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  2381.                         tindex2 = 2;

  2382.                         g->preflag = 1;

  2383.                     }

  2384.                 }

  2385. 

  2386.                 j = 0;

  2387.                 for(k=0;k<4;k++) {

  2388.                     n = lsf_nsf_table[tindex2][tindex][k];

  2389.                     sl = slen[k];

  2390.                     for(i=0;i<n;i++)

  2391.                         g->scale_factors[j++] = get_bitsz(&s->gb, sl);

  2392.                 }

  2393.                 /* XXX: should compute exact size */

  2394.                 for(;j<40;j++)

  2395.                     g->scale_factors[j] = 0;

  2396. #if defined(DEBUG)

  2397.                 {

  2398.                     printf("gr=%d ch=%d scale_factors:\n", 

  2399.                            gr, ch);

  2400.                     for(i=0;i<40;i++)

  2401.                         printf(" %d", g->scale_factors[i]);

  2402.                     printf("\n");

  2403.                 }

  2404. #endif

  2405.             }

  2406. 

  2407.             exponents_from_scale_factors(s, g, exponents);

  2408. 

  2409.             /* read Huffman coded residue */

  2410.             if (huffman_decode(s, g, exponents,

  2411.                                bits_pos + g->part2_3_length) < 0)

  2412.                 return -1;

  2413. #if defined(DEBUG)

  2414.             sample_dump(0, g->sb_hybrid, 576);

  2415. #endif

  2416. 

  2417.             /* skip extension bits */

  2418.             bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);

  2419.             if (bits_left < 0) {

  2420.                 dprintf("bits_left=%d\n", bits_left);

  2421.                 return -1;

  2422.             }

  2423.             while (bits_left >= 16) {

  2424.                 skip_bits(&s->gb, 16);

  2425.                 bits_left -= 16;

  2426.             }

  2427.             if (bits_left > 0)

  2428.                 skip_bits(&s->gb, bits_left);

  2429.         } /* ch */

  2430. 

  2431.         if (s->nb_channels == 2)

  2432.             compute_stereo(s, &granules[0][gr], &granules[1][gr]);

  2433. 

  2434.         for(ch=0;ch<s->nb_channels;ch++) {

  2435.             g = &granules[ch][gr];

  2436. 

  2437.             reorder_block(s, g);

  2438. #if defined(DEBUG)

  2439.             sample_dump(0, g->sb_hybrid, 576);

  2440. #endif

  2441.             s->compute_antialias(s, g);

  2442. #if defined(DEBUG)

  2443.             sample_dump(1, g->sb_hybrid, 576);

  2444. #endif

  2445.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); 

  2446. #if defined(DEBUG)

  2447.             sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);

  2448. #endif

  2449.         }

  2450.     } /* gr */

  2451.     return nb_granules * 18;

  2452. }

  2453. 

  2454. static int mp_decode_frame(MPADecodeContext *s, 

  2455.                            short *samples)

  2456. {

  2457.     int i, nb_frames, ch;

  2458.     short *samples_ptr;

  2459. 

  2460.     init_get_bits(&s->gb, s->inbuf + HEADER_SIZE, 

  2461.                   (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);

  2462.     

  2463.     /* skip error protection field */

  2464.     if (s->error_protection)

  2465.         get_bits(&s->gb, 16);

  2466. 

  2467.     dprintf("frame %d:\n", s->frame_count);

  2468.     switch(s->layer) {

  2469.     case 1:

  2470.         nb_frames = mp_decode_layer1(s);

  2471.         break;

  2472.     case 2:

  2473.         nb_frames = mp_decode_layer2(s);

  2474.         break;

  2475.     case 3:

  2476.     default:

  2477.         nb_frames = mp_decode_layer3(s);

  2478.         break;

  2479.     }

  2480. #if defined(DEBUG)

  2481.     for(i=0;i<nb_frames;i++) {

  2482.         for(ch=0;ch<s->nb_channels;ch++) {

  2483.             int j;

  2484.             printf("%d-%d:", i, ch);

  2485.             for(j=0;j<SBLIMIT;j++)

  2486.                 printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);

  2487.             printf("\n");

  2488.         }

  2489.     }

  2490. #endif

  2491.     /* apply the synthesis filter */

  2492.     for(ch=0;ch<s->nb_channels;ch++) {

  2493.         samples_ptr = samples + ch;

  2494.         for(i=0;i<nb_frames;i++) {

  2495.             synth_filter(s, ch, samples_ptr, s->nb_channels,

  2496.                          s->sb_samples[ch][i]);

  2497.             samples_ptr += 32 * s->nb_channels;

  2498.         }

  2499.     }

  2500. #ifdef DEBUG

  2501.     s->frame_count++;        

  2502. #endif

  2503.     return nb_frames * 32 * sizeof(short) * s->nb_channels;

  2504. }

  2505. 

  2506. static int decode_frame(AVCodecContext * avctx,

  2507. 			void *data, int *data_size,

  2508. 			uint8_t * buf, int buf_size)

  2509. {

  2510.     MPADecodeContext *s = avctx->priv_data;

  2511.     uint32_t header;

  2512.     uint8_t *buf_ptr;

  2513.     int len, out_size;

  2514.     short *out_samples = data;

  2515. 

  2516.     *data_size = 0;

  2517.     buf_ptr = buf;

  2518.     while (buf_size > 0) {

  2519. 	len = s->inbuf_ptr - s->inbuf;

  2520. 	if (s->frame_size == 0) {

  2521.             /* special case for next header for first frame in free

  2522.                format case (XXX: find a simpler method) */

  2523.             if (s->free_format_next_header != 0) {

  2524.                 s->inbuf[0] = s->free_format_next_header >> 24;

  2525.                 s->inbuf[1] = s->free_format_next_header >> 16;

  2526.                 s->inbuf[2] = s->free_format_next_header >> 8;

  2527.                 s->inbuf[3] = s->free_format_next_header;

  2528.                 s->inbuf_ptr = s->inbuf + 4;

  2529.                 s->free_format_next_header = 0;

  2530.                 goto got_header;

  2531.             }

  2532. 	    /* no header seen : find one. We need at least HEADER_SIZE

  2533.                bytes to parse it */

  2534. 	    len = HEADER_SIZE - len;

  2535. 	    if (len > buf_size)

  2536. 		len = buf_size;

  2537. 	    if (len > 0) {

  2538. 		memcpy(s->inbuf_ptr, buf_ptr, len);

  2539. 		buf_ptr += len;

  2540. 		buf_size -= len;

  2541. 		s->inbuf_ptr += len;

  2542. 	    }

  2543. 	    if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {

  2544.             got_header:

  2545. 		header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2546. 		    (s->inbuf[2] << 8) | s->inbuf[3];

  2547. 

  2548. 		if (check_header(header) < 0) {

  2549. 		    /* no sync found : move by one byte (inefficient, but simple!) */

  2550. 		    memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);

  2551. 		    s->inbuf_ptr--;

  2552.                     dprintf("skip %x\n", header);

  2553.                     /* reset free format frame size to give a chance

  2554.                        to get a new bitrate */

  2555.                     s->free_format_frame_size = 0;

  2556. 		} else {

  2557. 		    if (decode_header(s, header) == 1) {

  2558.                         /* free format: prepare to compute frame size */

  2559. 			s->frame_size = -1;

  2560.                     }

  2561.                     /* update codec info */

  2562.                     avctx->sample_rate = s->sample_rate;

  2563.                     avctx->channels = s->nb_channels;

  2564.                     avctx->bit_rate = s->bit_rate;

  2565.                     avctx->sub_id = s->layer;

  2566.                     switch(s->layer) {

  2567.                     case 1:

  2568.                         avctx->frame_size = 384;

  2569.                         break;

  2570.                     case 2:

  2571.                         avctx->frame_size = 1152;

  2572.                         break;

  2573.                     case 3:

  2574.                         if (s->lsf)

  2575.                             avctx->frame_size = 576;

  2576.                         else

  2577.                             avctx->frame_size = 1152;

  2578.                         break;

  2579.                     }

  2580. 		}

  2581. 	    }

  2582.         } else if (s->frame_size == -1) {

  2583.             /* free format : find next sync to compute frame size */

  2584. 	    len = MPA_MAX_CODED_FRAME_SIZE - len;

  2585. 	    if (len > buf_size)

  2586. 		len = buf_size;

  2587.             if (len == 0) {

  2588. 		/* frame too long: resync */

  2589.                 s->frame_size = 0;

  2590. 		memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);

  2591. 		s->inbuf_ptr--;

  2592.             } else {

  2593.                 uint8_t *p, *pend;

  2594.                 uint32_t header1;

  2595.                 int padding;

  2596. 

  2597.                 memcpy(s->inbuf_ptr, buf_ptr, len);

  2598.                 /* check for header */

  2599.                 p = s->inbuf_ptr - 3;

  2600.                 pend = s->inbuf_ptr + len - 4;

  2601.                 while (p <= pend) {

  2602.                     header = (p[0] << 24) | (p[1] << 16) |

  2603.                         (p[2] << 8) | p[3];

  2604.                     header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2605.                         (s->inbuf[2] << 8) | s->inbuf[3];

  2606.                     /* check with high probability that we have a

  2607.                        valid header */

  2608.                     if ((header & SAME_HEADER_MASK) ==

  2609.                         (header1 & SAME_HEADER_MASK)) {

  2610.                         /* header found: update pointers */

  2611.                         len = (p + 4) - s->inbuf_ptr;

  2612.                         buf_ptr += len;

  2613.                         buf_size -= len;

  2614.                         s->inbuf_ptr = p;

  2615.                         /* compute frame size */

  2616.                         s->free_format_next_header = header;

  2617.                         s->free_format_frame_size = s->inbuf_ptr - s->inbuf;

  2618.                         padding = (header1 >> 9) & 1;

  2619.                         if (s->layer == 1)

  2620.                             s->free_format_frame_size -= padding * 4;

  2621.                         else

  2622.                             s->free_format_frame_size -= padding;

  2623.                         dprintf("free frame size=%d padding=%d\n", 

  2624.                                 s->free_format_frame_size, padding);

  2625.                         decode_header(s, header1);

  2626.                         goto next_data;

  2627.                     }

  2628.                     p++;

  2629.                 }

  2630.                 /* not found: simply increase pointers */

  2631.                 buf_ptr += len;

  2632.                 s->inbuf_ptr += len;

  2633.                 buf_size -= len;

  2634.             }

  2635. 	} else if (len < s->frame_size) {

  2636.             if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)

  2637.                 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;

  2638. 	    len = s->frame_size - len;

  2639. 	    if (len > buf_size)

  2640. 		len = buf_size;

  2641. 	    memcpy(s->inbuf_ptr, buf_ptr, len);

  2642. 	    buf_ptr += len;

  2643. 	    s->inbuf_ptr += len;

  2644. 	    buf_size -= len;

  2645. 	}

  2646.     next_data:

  2647.         if (s->frame_size > 0 && 

  2648.             (s->inbuf_ptr - s->inbuf) >= s->frame_size) {

  2649.             if (avctx->parse_only) {

  2650.                 /* simply return the frame data */

  2651.                 *(uint8_t **)data = s->inbuf;

  2652.                 out_size = s->inbuf_ptr - s->inbuf;

  2653.             } else {

  2654.                 out_size = mp_decode_frame(s, out_samples);

  2655.             }

  2656. 	    s->inbuf_ptr = s->inbuf;

  2657. 	    s->frame_size = 0;

  2658. 	    *data_size = out_size;

  2659. 	    break;

  2660. 	}

  2661.     }

  2662.     return buf_ptr - buf;

  2663. }

  2664. 

  2665. AVCodec mp2_decoder =

  2666. {

  2667.     "mp2",

  2668.     CODEC_TYPE_AUDIO,

  2669.     CODEC_ID_MP2,

  2670.     sizeof(MPADecodeContext),

  2671.     decode_init,

  2672.     NULL,

  2673.     NULL,

  2674.     decode_frame,

  2675.     CODEC_CAP_PARSE_ONLY,

  2676. };

  2677. 

  2678. AVCodec mp3_decoder =

  2679. {

  2680.     "mp3",

  2681.     CODEC_TYPE_AUDIO,

  2682.     CODEC_ID_MP3,

  2683.     sizeof(MPADecodeContext),

  2684.     decode_init,

  2685.     NULL,

  2686.     NULL,

  2687.     decode_frame,

  2688.     CODEC_CAP_PARSE_ONLY,

  2689. };